Book Review: Cost-Benefit Revolution

22 minute read


Don’t let the dry reviews on the back cover deter you (Forbes says: “Clear, well-argued”). Cass Sunstein’s The Cost-Benefit Revolution is an awesome read for anyone interested in public policy or economics. He walks through the principles of cost-benefit analysis and considers a variety of challenges and case studies. The book won’t keep you at the edge of your seat (unless, you’re like, really into this), but Sunstein’s arguments are well-written and organized—a pleasure to consume.

Here’s the paperback and audiobook.

Utility: ⭐⭐⭐ (3/5)

Writing: ⭐⭐⭐ (3/5)

I took away five key ideas:

  • Cost-benefit analysis forces regulators to overcome biases and make a decision based on numbers.
  • Willingness-to-pay is an effective proxy for welfare, regardless of form. It can measure happiness, rights, meaning—even moral commitments.
  • Policymaking is messy. In contexts like mandatory labeling, national security, or free speech, cost-benefit analysis may not be appropriate.
  • Although regulators lack perfect knowledge, reforms can help them improve.
  • Courts evaluating regulations should defer to agencies, except when they make serious, arbitrary decisions.

Large parts of the book are related to work I did at NIST in the summer of 2020. I plan to reach out to Professor Sunstein for an interview—fingers crossed.

Here are my notes. I’m quite happy with how they turned out. I hope I’m improving.


The Triumph of the Technocrats

Executive agencies have undergone a cost-benefit revolution. It is founded on a simple principle:

No action may be taken unless the benefits justify the costs.

The principle itself doesn’t specify what the costs and benefits are. Nor does it specify who will do the accounting. But this abstract principle nonetheless shapes the way policies are evaluated.

It begins with the Administrative Procedures Act (APA), which required that regulations be non-arbitrary. President Reagan turned the APA into action. Executive Order 12291 outlined basic requirements to conduct cost-benefit analysis (CBA). One rule was that every regulation had to include a Regulatory Impact Analysis (RIA) that laid out the costs and benefits, both quantifiable and unquantifiable.

Reagan was eager about CBA for a couple reasons. First, the administration was filled with technocrats. Second, some officials thought CBA could curb regulatory overreach. Third, CBA could unify the missions of the numerous, disparate agencies. Finally, some advisors believed it would strengthen self-government by expanding presidential oversight of regulations.

Reagan’s order laid the foundation for modern CBA. George H. W. Bush didn’t touch the order. President Clinton’s follow-up order emphasized the importance of equity, but re-iterated the cost-benefit principle. President Obama reaffirmed the principle and went further by emphasizing the value of human dignity. Even Trump retained Obama’s order.

CBA is about welfarism, not utilitarianism. Although pleasure and pain (or your utility measure of choice) are relevant, CBA can incorporate any values, including rights, meaning, culture, and more.

A Foreign Language

Most people—average Joe’s and politicians alike—are strongly opposed to putting a economic value on life. The government does it anyways.

Populists want decisions to be made through deliberation. Technocrats look instead to numbers, drawing from science, statistics, and economics; they think the government should do what is best, regardless of whether the public agrees.

One reason to support the technocrats is that everyday people are susceptible to biases. The availability heuristic causes people to overweight recent personal experiences in evaluating risk. Individual biases are compounded by informational cascades, where people will adopt beliefs just because others do so. Combined, the two factors explain the overblown reactions to Ebola, GMOs, and nuclear energy.

People also struggle to evaluate complex systems. In simulations, people can’t foresee how their decisions may interact and backfire. For example, a living wage may hurt the poor on net by raising unemployment. CBA looks at the entire system.

When people use a foreign language, they often make fewer of these mistakes. For example, when Americans consider a bet in French, they think in terms of numbers are less loss averse.

CBA functions like a foreign language. It’s a way to reframe problems in terms of the underlying costs and benefits; in doing so, it mitigates biases.

Willingness to Pay and the Value to Life

CBAs rely on the value of a statistical life (VSL), which is usually around $9 million. VSL does not refer any particular life; it’s about statistical risks of death. Specially, it evaluates willingness to pay (WTP) for reduced risks.

In Easy Cases, people who benefit from a regulation pay all of the cost. For example, consumers may pay a slightly higher price for goods in exchange for a uniform improvement in air quality.

In Hard Cases, beneficiaries pay only some of the cost. For example, a workplace safety regulation could primarily benefit workers, but end up costing consumers.

In harder cases, a regulation may be justified on distributive grounds. Workers could be much worse off than consumers, and so even though consumers lose more than workers gain, a regulation could still produce overall welfare. Generally though, inequalities should be rectified by a distributive income tax.

There are two ways estimate VSL. The first is to look at real-world markets, like job markets for risky professions. These tend to be noisy, since although a construction job involves more risk, people may choose a job for a variety of reasons.

The second way is contingent valuation studies, which ask people how much they are willing to pay to reduce risks. While these studies isolate the value of mortality risks, they present strange, hypothetical scenarios and usually receive a broad range of answers.

WTP is usually an effective proxy for welfare. The government should not force people to spend more than they are willing to pay to reduce a given risk—doing so would inflict a net harm. In this sense, WTP also respects autonomy by recognizing the importance of peoples’ preference.

Although VSL is often calculated across an entire population, different people can have different VSLs. In general, poor people have a lower VSL than rich people. This is a good thing: not because they are less valuable than rich people, but because they shouldn’t be forced to pay more than they are willing to pay for reduced risks. A uniform VSL ends up costing the poor.

In practice, disaggregating WTP across groups may require unavailable information. It also relies on questionable contingent valuation studies. Often, it’s not even possible to tailor policies towards subgroups.

WTP faces several potential objections. First, people may be miswanting: they claim to want things that don’t actually increase their welfare. A similar effect is adaptive preferences, where people grow accustomed to health and environmental risks and lose interest in fixing them. Both of these objections, however, usually aren’t relevant for low-probability mortality risks.

A second objection relates to behavioral market failures. People suffer systematic errors: present bias causes excessive discounting of future risks; over-optimism leads people to underestimate risk; people can’t comprehend low-probability events, like distinguishing a 1/100,000 event from a 1/1,000,000 event. In practice, these errors are mitigated when aggregated across large populations. Moreover, if the effect is significant, analysts can correct for them afterwards.

A third objection is based on rights. If someone is at great risk of dying from water pollution, then they should be protected, even if they are personally unwilling to to pay for it. Some things may be truly “off limits” to CBA, such as racial discrimination, workplace sexual harassment, and animal cruelty. But the risks are usually very small. In addition, the correct response to people undervaluing their own safety is to provide a subsidy.

A final objection is that WTP doesn’t account for third-party effects. When someone dies, their family, friends, and community may also suffer; the preferences of these third parties are not accounted for by WTP. This is worth exploring further.

Hard cases provide a different challenge. A policy with net monetary benefits could have net welfare costs, and vice versa. On top of that, poor people may gain a lot more than the rich from the same welfare boost. Situations like these lend support to prioritarainism: the idea that we should increase overall welfare with priority given to the least advantaged. Using a unitary (rather than disaggregated) VSL could be justified in instances where the poor have a lot to gain and little to lose. If poor people are not footing the bill, then the autonomy case for WTP is also weakened.

Some people may suggest that utilitarianism is a better way to increase welfare. In reality, however, rules that pass CBA but fail a utilitarian test are rare. Often, the VSL benefits massively outweigh the costs. Regulators tend to avoid distributional effects because they are hard to quantify and are subject to interest-group influence. The best approach in Hard Cases is to consider distributional effects only when appropriate.

Welfare: The Master Value

Consider a requirement that cars have back cameras. A lot of factors aren’t easily monetized. First, those who die in back-over crashes tend to be young children with many years of life ahead. Second, a parent who kills their own child undergoes extensive, unbearable grief. Third, the cost of the rule is distributed across an entire population. Fourth, the cameras make driving more pleasurable.

There are several proposed ways to measure welfare. “Evaluative” welfare—which is about overall life satisfaction—is a proxy for overall life aspirations, although the measure is subject to cross-national differences and strange biases. “Experienced” welfare—which asks how people feel about specific activities—is proxy for emotional states.

Someone could object to these measures by arguing that humans care about more than just happiness. Surely, we care about respect, knowledge, beauty, and more. But this is the advantage of CBA—if people value these non-hedonist goods, then they should be willing to pay for them too.

Even good measures of welfare are hard to apply to specific policies. People who have tried have reached very unintuitive valuations. Trying to directly measure welfare requires way too much guesswork in extrapolating results and interpreting numbers.

Future research may provide avenues to directly measure and consider welfare effects. For now though, WTP is the best proxy for welfare.

The Knowledge Problem

Friedrich Hayek distrusted socialism because of a the knowledge problem: relevant facts about the consequences of a policy are dispersed throughout society, so a central planner could never know enough to successfully plan an economy. That is to say, regulators will lack the knowledge to make accurate assessments of costs and benefits.

While Hayek’s claim made sense for his era, modern regulators have access to fine-grained details about swaths of economic actors. There are four further reforms that can fill these gaps in knowledge.

The first is notice-and-comment rulemaking. First established by FDR and later operationalized by Obama, online platforms enable the public to comment on proposed regulation. Although the process risks benefitting interest groups, it let’s officials draw insights from everyday people.

The second reform is retrospective analysis. Conducting a CBA on an existing regulation can determine whether it should stay in place. It can also verify the accuracy of the initial CBA. The handful of studies comparing ex-ante and ex-post CBAs are small and identify no systemic errors.

A third reform is rapid learning. Randomized controlled trials are the strongest kind of evidence. Regulators could, for example, test the effect of banning cell phones, requiring rear view cameras, or adding fuel economy labels. In the UK, the Nudge Unit has spearheaded randomized controlled trials in many policy contexts.

A fourth reform is measure-and-react. In the fast-paced environment of Silicon Valley, companies experiment with different products and adjust them in real-time as they receive feedback. Governments could do the same thing in responsive environments like airport security or traffic control.

Moral Commitments

WTP can measure the strength of moral commitments. When people feel strongly about an issue—whether it be dolphins, rape victims, or their children—they lose personal welfare so long as the issue persists. The EPA has used proxies to value ecological protection, and courts have even required it.

The problem is that even if we decide to address an issue—say, for example, regulating blood diamonds—it doesn’t tell us how stringent the regulation should be. Setting the appropriate level of regulation is a matter of CBA, and valuing moral commitments can be an effective tool.

Of course, the moral commitments of racists and homophobes are not worth evaluating. Constitutionality provides a side constraint: CBA should not evaluate moral commitments that violate the Constitution. WTP for assault, torture, or other horrific outcomes should also be ignored.

In practice, agencies could conduct contingent-valuation studies on moral commitments. WTP may be skewed, since there are resource tradeoffs (saving the dolphins can district from saving the children). But in principle, the point remains that moral commitments should count in CBA.

On Mandatory Labeling

Incomplete information is a product of market failure. Consumers don’t have the technical background to request information like nutrition facts. Information is also a public good, so mandatory labeling faces a collective action problem. Moreover, when labeling benefits third parties—as in the case of secondhand smoke warnings or vegan labels—consumers might not be motivated to act.

Unraveling mitigates the problem. If consumers prefer non-GMO foods, then organic grocers have an incentive to label their foods as non-GMO. But it isn’t enough. First, companies could mislabel their products or provide non-standardized information. Second, it doesn’t do much for third parties. Third, consumers have limited attention; they might not infer from the lack of a ‘non-GMO’ label that a product is genetically modified.

There are four approaches to CBA for disclosure requirements. The first is to simply admit that there isn’t enough information quantify benefits—not great.

The second is to use breakeven analysis, where an agency asks: “what level of benefits would justify the costs?” For example, for a fuel economy disclosure requirement that costs $15 million, an agency may calculate that the average car purchaser must receive at least $1 in benefits. It still involves some guesswork.

A third approach is to specify outcomes in terms of economic goods and health. A final approach is to directly measure WTP.

Regulators overlook several sources of costs. The first is a cognitive tax. More information diverts consumer attention. The second is a hedonic tax. People who quit smoking or eat healthier miss out on their old pleasures. The unlucky folks who can’t quit will feel even worse. The third is welfare loss. If someone switches to a fuel efficient car, they may be losing out on other nice features.

For benefits, WTP would be ideal. Unfortunately, present bias and optimistic bias depress WTP for disclosure. Directly assessing endpoints is also difficult and tends to underweight the costs outlined above.

GMOs are an interesting example. According the scientists, they improve some health outcomes, don’t post any health risks, and carry only small environmental risks. Yet, public opinion is decisively against GMOs. In surveys, the most adamant opponents don’t even care about costs and benefits.

Evaluating the benefits of a mandatory ‘GMO’ label is hard. Although people may be willing to pay a small amount, it probably wouldn’t lead a concrete benefit. Moreover, any WTP measure would be unreliable and biased. The best choice may be to use breakeven analysis with consumer preferences and potential environmental harms.

The Precautionary Principle is relevant here: if a policy could cause severe harm, it should be avoided absent certainty about its safety. Irreversibility is especially important. It seems logical to place special weight on harms that cannot be corrected. The principle would suggest that a product with modest benefits but a risk of catastrophic harm should nonetheless be banned.

The Role of Courts

Courts should be cautious in reviewing CBAs because they lack technical expertise. Some cases obviously violate the APA (which bans arbitrary regulations), such as refusing to quantify costs or passing a policy despite massive net costs. In many other cases, the courts should defer to agencies.

Two decisions are crucial. First, Michigan v. EPA establishes that agencies should consider economic factors if possible; otherwise, it’s unlawful under the APA. Second, in Business Roundtable, the D.C. Court of Appeals mandates the SEC to quantify costs and benefits, without specifying when agencies may deem quantification implausible.

There are some nuances. If Congress requires an agency to ignore CBA, then courts must respect that statute. Agencies that make strange accounting decisions—for example, failing to discount costs or benefits—could be acting arbitrarily. At the same time, agencies probably don’t have to analyze alternatives with the same standards.

The court has recognized situations where quantification is unreasonable. When that happens, agencies may simply go ahead with a rule if the stakes are low. It could also make rough estimates that justify a decision, even in absent of numbers. Additionally, when costs and benefits are similar, fairness and dignity may tip the balance.

Privacy and National Security

Since 9/11, two groups of pundits have arisen. The first believes that national security must come first and that high-magnitude terrorist threats should be prioritized. Another group values privacy above all else and warns of the risk of despotism.

Surveillance can combat terrorism, but it could also hurt privacy and civil rights. In contexts like these, the Precautionary Principle is useless. The precautionary steps—such as monitoring cell phone data—can themselves post serious risks. Appealing to worst case analysis is unhelpful.

Yet, people use it anyway, often because of biases. Motivated reasoning may cause zero-in on either terrorism or privacy. They could also be subject to the availability heuristic, loss aversion, or probability neglect.

One approach may be to consider the worst case scenarios for both options. A massive terrorist attack is probably worse. Breakeven analysis could also be helpful. How much would people be willing to pay for a 10% reduction in risk of a terrorist attack that costs $200 billion?

People may compromise on gratuitous costs. For example, making the government explicitly request metadata could help protect privacy without impeding on national security. On the flip side, innocent invasions of privacy may be justified to fight terrorists.

Free Speech

Federal regulators evaluate costs and benefits in terms of expected value: if an event happens in the future or has only a small chance of happening, regulators will discount it and include it in the analysis.

Modern free speech laws work differently. Obviously, misuses of speech such as incitement of violence should be regulated. Originally, the decision to regulate speech was governed by something like CBA, with a precedent established in Dennis v. United States. But Brandenburg v. Ohio, in a decision spearheaded by Justice Brandeis, set the bar much higher.

The new standard is the “clear and present danger” test: speech is only incitement if the foreseen violence is imminent and likely. Drawing on J.S. Mill, Brandeis argued that if the harm was not imminent, then counterspeech—people disagreeing with the inciter—should be able to combat it.

The justification is shaky. If, for example, a speaker risks causing 200 deaths over the following two years, then it seems justifiable to act now, even amid uncertainty and lack of imminent threat. Moreover, immediate action is often better, since staling could mean that it becomes too late.

There are three plausible defenses of the clear and present danger test. First, cost and benefits are hard to quantify for speech. This objection is reasonable—free speech isn’t exactly a commodity, and willingness to pay doesn’t make much sense when the listeners are the primary beneficiary. But we don’t need economic analysis. A serious cost like a possible terrorist attack can still be understood and compared to the benefits of speech.

A better objection is that governments have institutional biases. Since public officials benefit from greater security, they have a tendency to overstate the risks of speech. CBA might allow officials to overregulate speech under the guise of a neutral standard. The clear and present danger doesn’t obviously solve this, but it may be a better alternative.

The best objection is that, for most of history, the clear and present danger test works in practice. Circumstances where people seriously threaten to revolt or kill police officers are far and few between. At least for the twentieth century, it’s hard to find a scenario where a CBA would have done better than the imminence requirement. Of course, in an era where the risk of international terrorism is real, this environment may soon change.

Conclusion: Best-Laid Plans

CBA is not a political issue. When the Obama restriction rejected on environment regulation but simultaneously adopted another, both environmentalists and business groups where confused. They though it was an ideological issue. It wasn’t. It was a question of costs and benefits.

Democrats are not angels. Following the Montreal Protocol, the Obama administration banned MaxAir—an asthma inhaler used by hundreds of thousands—without considering the massive costs and miniscule benefits. CBA unequivocally defends welfare, regardless of who is in office.

Trump’s current regulatory regime, “one in, two out,” risks complicating the basic role of CBA. The right approach is to thoroughly evaluate new rules and to scrutinize the necessity of existing rules. If Trump fails to pass policies with great costs but even greater benefits, then the administration will set back the cost-benefit revolution.

Perhaps what we need is an new institution. Reagan’s OIRA and the Domestic Policy Council are effective overseers, but a truly independent institution devoted to vetting significant regulations could be a very good idea.

There are two challenges moving forward. First, monetary figures are not a perfect proxy for welfare, especially in the context of unemployment effects, Hard Cases, and distributed costs. Second, agencies face numerous knowledge problems, whether in making mistakes, misjudging magnitude, or even fudging numbers. Challenges like these manifested in the ACA’s electronic health record requirements: the rules cause significant stress of doctors, and probably didn’t lead to many benefits.

The future is bright. Promising solutions to the knowledge problem include public comment, retrospective review, and randomized controlled trials. Measure-and-react is especially promising in long-term environments. CBA has come a long way since the Reagan era, and it has contributed immensely to welfare. The revolution is unfinished.