I. What Is Efficiency?

Depending on the speaker, “efficiency” can denote diverse and contradictory concepts. One common definition is Pareto efficiency. A state of affairs is Pareto efficient if no change could increase one individual’s well-being without decreasing the well-being of another individual. But the Pareto criterion is a relatively harsh one: even many appealing policies will harm someone. For example, a regulation that prohibits a factory owner from dumping toxic waste into a lake will not be a Pareto improvement because it harms the factory owner, no matter how toxic the waste or how culpable the owner.

Kaldor-Hicks efficiency relaxes this rule by introducing a compensation assumption. It asks whether a change would improve the well-being of some and not harm any others, assuming that the winners were to compensate the losers. Because Kaldor-Hicks deals with hypothetical compensation, rather than actual compensation, Kaldor-Hicks improvements are much more common. The toxic waste regulation might be a Kaldor-Hicks improvement but not a Pareto improvement, if everyone would be at least as well off were the polluted-upon to compensate the polluter.Footnote ¹

Neither Pareto nor Kaldor-Hicks analysis identifies unique efficient policies; many competing policies may each be “efficient” in a Pareto or Kaldor-Hicks sense,Footnote ² or two different policy options may be incomparable.Footnote ³ Pareto and Kaldor-Hicks simply attempt to identify some set of policies as efficient without providing a way to discriminate in every case. Despite these limitations, many economists prefer Pareto and Kaldor-Hicks to the maximization of social welfare because they do not explicitly require cardinal or interpersonal comparisons of utility—they only require each individual to identify which of two situations she prefers, without specifying by how much.

Wealth maximization is one attempt to implement a Kaldor-Hicks criterion in practice. Wealth maximization imagines that for any proposed policy change, each individual would pay a certain amount for the change to occur (willingness to pay) or accept a certain amount to forgo the change (willingness to accept). We could then sum all of the willingnesses to pay or willingnesses to accept, and if the amount were positive, then the policy change would be wealth maximizing.Footnote ⁴

Kaldor-Hicks efficiency and wealth maximization are especially important because they motivate real-world cost-benefit analysis, which attempts to quantify costs and benefits in terms of willingness to pay or accept. Although some scholars conflate wealth maximization with Kaldor-Hicks analysis,Footnote ⁵ they differ in their formal definitions, and wealth maximization encounters some particular problems that Kaldor-Hicks does not.Footnote ⁶

A final approach to efficiency focuses on “deadweight loss” or “excess burden.” This approach concerns welfare losses from government policies that “distort” market behavior. For example, a tax might cause a “loss of consumer (and producer) surplus when comparing the after-tax world to the before-tax world.”Footnote ⁷ Deadweight loss analysis imagines that there is an efficient, undistorted, pre-tax world, and calculates the social cost of moving from that world to a less efficient, distorted, post-tax world.

Between these four concepts, Kaldor-Hicks efficiency is generally what law and economics scholars mean when they discuss efficiency simpliciter. Deadweight loss receives greater attention in some specialized fields, like taxation and antitrust. The distinctions matter: each of these definitions has its advantages and disadvantages, as we will see below.

How is efficiency used in the literature? Scholars frequently contrast efficiency with redistribution—Arthur Okun described the choice between the two as the “big tradeoff” in public policy.Footnote ⁸ In legal scholarship, commenters have disagreed on whether government policies should be selected only with efficiency in mind, or with distributional effects in mind as well. In tort law, for example, we might expect that trespassers are systematically poorer than landowners, or that owners of pollution-generating factories are systematically richer than their polluted-upon neighbors. Louis Kaplow and Steven Shavell famously argued that legal rules should be chosen only on efficiency grounds, because it is systematically better to redistribute through taxes and transfers.Footnote ⁹

Critics have attacked Kaplow and Shavell’s argument since it was first introduced. Some scholars have argued that legal rules are efficient avenues for redistribution and provided economically grounded explanations, including both behavioralFootnote ¹⁰ and neoclassical explanations.Footnote ¹¹ Other scholars have argued that a preference for redistribution through taxes and transfers in a first-best world is irrelevant, because taxes and transfers are inadequate in our n ^th-best world.Footnote ¹² Still others have suggested that the assumptions undergirding Kaplow and Shavell’s theoretical model are unrealistic.Footnote ¹³ Many of these arguments cabin the scope of efficiency analysis in compelling and useful ways.

But a more recent line of scholarship advances an even more fundamental rebuttal. These articles cast doubt on the concept of efficiency itself—they argue that efficiency is incoherent or biased. They therefore represent a more ambitious project, not just to erode the priority of tax-and-transfer at the margins but to jettison it entirely.

I argue that these arguments all reduce to familiar disputes over the empirical assumptions underlying efficiency analysis, and that efficiency is neither biased nor incoherent. While many of the objections to specific uses of efficiency analysis are valid, they should be addressed on their own terms and not through broad condemnations of the concept of efficiency.

II. Why Efficiency Is Not Incoherent

One prominent critique of efficiency is that it is simply incoherent—that efficiency contains so many problematic assumptions that it is meaningless. Neil Buchanan and Michael Dorf have recently made this point with particular persuasiveness.

Here is the argument: efficiency must be analyzed with respect to a baseline of property rights. However, property rights are themselves the products of legal rules.Footnote ¹⁴ The selection of optimal property rights will therefore change the optimal set of other legal rules; and any changes to the optimal set of other legal rules will change the optimal set of property rights. Thus, efficiency is circular. In the absence of some “natural” baseline set of legal rules, it is impossible to assess which legal rules maximize efficiency.Footnote ¹⁵

Consider a specific example that Buchanan and Dorf provide: sport-utility vehicles (SUVs). They argue that SUVs exist as a class of vehicle in the United States primarily to meet federal fuel efficiency standards from the 1980s. It is therefore incoherent to ask what the “efficient” quantity of SUVs would be. “The supply curve only exists at all because the government made a choice decades ago that it did not have to make (but for which there was no default baseline).”Footnote ¹⁶

Buchanan and Dorf seem to suggest that supply and demand cannot be considered at all when designing legal rules. But the alternatives hardly seem appealing: pure formalism, perhaps, or rules selected solely on the basis of their distributional effects, eschewing any consideration of the rules’ incentive effects. Thus, Buchanan and Dorf prove too much—even critics of efficiency generally acknowledge that incentive effects matter, even if they should not be the sole considerations in crafting legal rules.

Moreover, not every type of efficiency analysis requires a baseline. Buchanan and Dorf purport to discuss Pareto and Kaldor-Hicks efficiency, as is standard in law and economics.Footnote ¹⁷ But their argument seems most apt as a critique of deadweight loss. Of the types of efficiency discussed above, Pareto and Kaldor-Hicks efficiency do not require a baseline; only deadweight loss analysis and wealth maximization do, for different reasons.

Pareto requires only that we compare each individual’s utility in a world with and without SUVs, which is a conceptually feasible task. Kaldor-Hicks is somewhat more complicated because it requires us to imagine compensation is paid. However, the compensation can be paid in whatever goods exist in the particular hypothetical world in question; in SUVs, sedans, bananas, dollars, or anything else. Hypothetical compensation does not need to be paid in any particular coin. Thus, again, Kaldor-Hicks efficiency requires no baseline and avoids charges of incoherency.Footnote ¹⁸

Put differently, Pareto and Kaldor-Hicks require us to compare two different worlds with comprehensively specified packages of background property rights and other legal rules, and then compare individuals’ welfare levels between these two worlds. Pareto and Kaldor-Hicks are inherently relational and thus require no uniquely privileged comparison world to make assessments of efficiency.

Buchanan and Dorf’s argument is most apt with deadweight loss analysis, focusing specifically on supply, demand, and the deadweight loss “triangle.” This formula does, indeed, suffer from a baseline problem. And so does wealth maximization, albeit for reasons that Buchanan and Dorf do not discuss.

First, wealth maximization is not well-defined because wealth is not well-defined between different states of the world. For example, Rolls-Royces might be more expensive, and Toyotas cheaper, in a society with more wealth inequality. Changes to entitlement regimes therefore change relative prices, so that money is not a consistent numeraire, and wealth in one hypothetical world cannot be directly compared with wealth in another.Footnote ¹⁹ This problem has been widely known and discussed among economists for decades.Footnote ²⁰ It may, however, be relatively insignificant in practice, since most policies will have little impact on overall relative prices.

Relatedly, wealth maximization requires a baseline in the sense that individuals’ willingness to accept or pay is determined by their existing endowments. As Duncan Kennedy has explained, if two individuals are stranded in the desert with a life-saving glass of water, wealth maximization will leave the glass with whichever individual started with it, because no amount of money could persuade the glass’s current possessor to give it up. These “wealth effects” make wealth maximization dependent on the choice of baseline and, therefore, arbitrary.Footnote ²¹

If deadweight loss minimization or wealth maximization were the only goal, these problems would be insuperable—which is one of many reasons why they are not the only goal. Richard Posner’s onetime suggestion that wealth maximization might be an end in itself was roundly criticized within the legal academy, with Posner eventually conceding that wealth maximization is merely a pragmatic means to the ultimate end of welfare maximization.Footnote ²² Among legal economists in general, welfare maximization is the standard fundamental justification for wealth maximization and cost-benefit analysis.Footnote ²³

Crucially, even if deadweight loss minimization and wealth maximization are not globally well-defined, they may still be locally well-defined; that is, we might still identify which moves minimize deadweight loss or maximize wealth holding all other factors constant. Under this approach, all background legal rules, other than the specific legal rules under consideration, are assumed as a baseline. This simplifies analysis but comes with a cost—it may make deadweight loss minimization and wealth maximization imperfect proxies for welfare maximization, if that is the ultimate goal. But using these assumptions, deadweight loss and wealth maximization remain at least coherent concepts as used in practice.

Consider again Buchanan and Dorf’s SUV example. Imagine that the National Highway Traffic Safety Administration (NHTSA) is considering whether to require electronic anti-rollover protection, to prevent a kind of accident to which SUVs are especially prone. The NHTSA conducts conventional cost-benefit analysis and estimates that the benefits of the regulation would exceed its costs—in other words, that the regulation is wealth maximizing, or that it minimizes deadweight loss. The cost-benefit analysis is necessarily a ceteris paribus exercise—it holds all other factors constant, including the existence of SUVs and by extension the existence of the regulatory environment that encourages consumers to buy SUVs.

This may or may not be the best method to use, but it is at least a well-defined, unambiguous method. Costs and benefits are unambiguously quantifiable within this framework—the calculations will involve assumptions, but the assumptions are definite and tractable. If the regulations encouraging consumers to purchase SUVs are not wealth-maximizing, then under the tax-and-transfer paradigm, those regulations should be removed. And if they are removed, the cost-benefit analysis regarding anti-rollover protection would become superfluous as well. Wealth maximization is not a single, global analysis. In theory and in practice, it is a series of piecemeal local assessments.

Again, neither wealth maximization nor deadweight loss minimization are ends in themselves. They are merely techniques that are instrumentally useful in attempting to reach some other goal, welfare maximization being the usual candidate. If these tools do not maximize welfare, they should be discarded—and there are many reasons to believe that they may not in fact be welfare-maximizing, especially if society inadequately redistributes through taxes and transfers. (As noted above, the general theoretical argument that we should maximize regulatory efficiency and then redistribute elsewhere has come up against legitimate criticism on empirical grounds.) However, this is a step away from efficiency being incoherent, and our doubts about the efficacy of efficiency should be confronted and discussed substantively rather than assuming them away.

In fact, Pareto and Kaldor-Hicks arguably also operate better as means rather than ends. Pareto and Kaldor-Hicks efficiency do not identify unique optima—many different policies might be Pareto or Kaldor-Hicks efficient in the sense that there are no available Pareto or Kaldor-Hicks improvements. Thus, we need some other criterion (like welfare maximization) to break the tie between these many possible “efficient” policies.Footnote ²⁴

If efficiency analysis, in whatever form, does not satisfy this other criterion, it should be discarded. But the question whether it does in fact satisfy this criterion is also a cognizable question. Of course, efficiency is complicated, and the calculation process is not easy. It is assuredly difficult; but it is not incoherent.

III. Why Efficiency Is Not Biased

Zachary Liscow advances another influential argument against efficiency by contending that it is generally “rich-biased” in a manner that will “give disproportionate legal entitlements to the rich for free, exacerbating inequality.”Footnote ²⁵ In Liscow’s view, any analysis (like cost-benefit analysis) that uses willingness to pay as its measure of efficiency will generally favor the rich—the rich will generally be willing to pay more, because they have more money. (Other authors have made similar arguments, including Edwin Baker, Matthew Adler, and Eric Posner,Footnote ²⁶ but Liscow’s is a particularly recent and powerful rendition of the critique.)

Which version of efficiency does Liscow discuss? His discussion and mathematical proofs suggest he has in mind what I call “wealth maximization,” although he calls it “Kaldor-Hicks efficiency.” The move to equate Kaldor-Hicks with wealth maximization has precedents in law and economics, and is not a fatal flaw in any case—wealth maximization is one attempt to instantiate Kaldor-Hicks, subject to the issues with relative prices and wealth effects noted above, which are incidental to Liscow’s analysis.

In a neat piece of analytical symmetry, Liscow’s argument also assumes the existence of a neutral baseline. He establishes the baseline by noting that “[a]n efficient policy is neutral if, as one gets richer, efficient legal rules do not change one’s legal entitlements.”Footnote ²⁷ In contrast, a policy “is poor-biased if, as one gets richer, one gets fewer legal entitlements from efficient legal policies.”Footnote ²⁸ And a policy “is rich-biased if, as one gets richer, one tends to get more legal entitlements from efficient policies.”Footnote ²⁹ He notes as an example that “everyone has the same willingness to pay for one dollar in increased or decreased income.”Footnote ³⁰ Thus, a policy that provided every individual with a flat amount of money would be neutral. In contrast, a policy that tended to provide richer citizens with more money would be rich-biased.

There are several problems with this choice of baseline. First, it only considers public spending, without considering sources of public revenue. This is a classic trope in tax policy: the distributional effect of public spending cannot be properly analyzed without understanding how that spending is funded, and vice versa.Footnote ³¹

For example, imagine a new demogrant that gave each citizen one dollar. This policy would be neutral within Liscow’s framework, neither rich-biased nor poor-biased. But the actual distributional effect of the demogrant depends on how it is to be funded. If the demogrant were funded by our federal income tax—which is generally progressive, so that the rich pay a larger percentage of their incomes than the poor—then such a demogrant would substantially redistribute from the rich to the poor. In fact, the tax rate would not even need to be progressive in order to effect redistribution. It is a well-known result within the literature on universal basic income that a demogrant plus a flat tax can effect substantial redistribution.Footnote ³²

To take the strongest version of Liscow’s argument, consider one of his examples: public parks. A government that maximizes wealth outside of tax policy will build parks in line with residents’ willingness to pay rather than in line with the utility that residents actually derive from the parks.Footnote ³³ Because rich residents are able to pay more, this means that the government will in general spend more on parks for the rich than parks for the poor. In Liscow’s terms, this policy is therefore rich-biased.

But this characterization hangs on Liscow’s definition of neutrality, which crucially includes only spending and not taxing. Once taxes are taken into account, the park system might significantly redistribute from the rich to the poor. This would be the case if public works were generally funded through progressive taxes. To put it more concretely, imagine that the National Parks Service is deciding whether to fund a new national park, using the proceeds of federal income taxes. The average taxpayer within the bottom 50 percent of the income distribution paid approximately $694 in federal income taxes in 2017. The average taxpayer within the top 10 percent of the income distribution paid approximately $132,168 in federal income taxes in 2017.Footnote ³⁴ It might be that the new national park would benefit the rich more than the poor—but would it benefit them 190 times more?Footnote ³⁵

This alternative baseline emphasizes the actual source of funds and the actual alternative to contemplated policies. If the Parks Service decided not to fund any parks, would the poor taxpayer benefit from the elimination of this “rich-biased” policy? If she bore little of the cost of the parks, she would likely prefer to keep them.

The essential distinction between my alternative baseline and Liscow’s baseline is that mine assumes that public spending is funded through tax revenues. Moreover, it assumes that this is also true on the margins—that any contemplated additional spending will generally be funded through taxation with roughly the same distributional profile. Liscow instead assumes that distributional effects are “sticky,” such that a regressive policy generally will not be offset by any change to the tax system.Footnote ³⁶ This is likely true when considering policies in isolation: no congressperson would observe an Environmental Protection Agency (EPA) regulation that benefits the rich and conclude that we must therefore raise the top tax rate by 0.001 percent. But on a wider scale, this assumption becomes difficult to defend. New public spending will ultimately be paid for somehow; and as a rough approximation, it seems most reasonable to assume that it will be paid for in the same manner as existing public spending.Footnote ³⁷

Under Liscow’s baseline, rich-biased policies are exceedingly common; but taking the source of funds into account, rich-biased policies become exceedingly rare. (Most taxes in the United States are progressive; sales taxes are a key exception, but they make up a relatively small percentage of overall government revenue, and there is no national general sales tax.Footnote ³⁸) If tax rates were merely flat, a policy would be distributionally neutral if willingness to pay increased as a simple linear function of income. But so long as taxes are progressive, policies will only be distributionally rich-biased if the rich are willing to pay more as a percentage of their income, and if their willingness to pay scales-up more quickly than rates of tax. Thus the dramatic difference between average tax revenues from the rich and poor taxpayer mentioned above. The rich taxpayer pays an average federal income tax of 21.5 percent, whereas the poor taxpayer pays an average federal income tax of 4 percent. This is compounded by the fact that the rich taxpayer has a substantially higher income, thus the result that the rich taxpayer pays 190 times more than the poor one.

The rich might value some public goods more as a percentage of their income, compared to poor taxpayers. Poor taxpayers can reap higher marginal benefits from spending on private goods—at the margin, they are more likely to spend on necessities, while the rich might spend on luxuries. Having satiated her interest in personal material comforts, the rich taxpayer might conceivably be willing to pay for some public goods a percentage of her income 5 times greater than the poor taxpayer’s percentage, which corresponds to an amount 190 times greater in raw expenditure.Footnote ³⁹ But it is not obvious what public goods this would apply to. Moreover, it seems the exception rather than the rule, since many of the most important government-provided public goods—national defense, clean air and water—are basic necessities rather than luxuries. After taking both spending and taxing into account, these policies would seem to favor the poor, even though they meet Liscow’s criterion for rich-bias.

I am not arguing that present policy should be altered to favor the rich. I only mean to say that efficiency can be meaningfully described as poor-biased just as easily as it can be described as rich-biased. The finding of bias is thus heavily dependent on the neutral baseline that one chooses, and there is reason to doubt Liscow’s choice.

But what if legal rules are in question rather than public spending? In these cases, there is no issue about the source of funds. Liscow tends to emphasize rich-bias in public spending,Footnote ⁴⁰ presumably because these cases are intuitive and common. And he acknowledges that, “[i]n practice, torts typically do not offer any relief to those whose quality of life is harmed by worse health but who suffer no financial harm, such as compensation for pain and suffering, in ways that give more compensation to the rich than to the poor.”Footnote ⁴¹ However, it seems likely that all legal rules that protect property in a broad fashion are systematically rich-biased under Liscow’s criterion. This is so because the rich tend to have more property than the poor, so they will benefit more from such rules in dollar terms.

One hypothetical example of a neutral or poor-biased property rule would be one written narrowly to protect the sort of property that the poor disproportionately own. So we could, say, enact theft prevention laws that protect mopeds but not Rolls Royces. Alternatively, we could effect redistribution within legal proceedings by explicitly changing standards of proof or levels of damages to benefit poorer litigants. For example, poor plaintiffs might be required only to meet the preponderance-of-evidence standard in civil proceedings (the standard under current law), whereas rich plaintiffs might be required to meet the beyond-a-reasonable-doubt standard.Footnote ⁴²

So, within Liscow’s framework, a tort rule that broadly requires arsonists to pay compensation for property they destroy is rich-biased; tweaking the rule to benefit poor arsonists and disadvantage rich ones could render it neutral.Footnote ⁴³ But this is a highly counterintuitive understanding of legal neutrality. In the examples above, rich-bias seems more akin to what most people would consider a neutral baseline, given that the legal rule was not selected to benefit the rich and applies in a nondiscriminatory fashion.

Again, I am not arguing that we adopt an alternative neutral baseline. I instead question whether we should frame our analysis in terms of rich-bias or poor-bias at all. Neutrality is important in Liscow’s framework because neutral wealth-maximizing policies are presumptively acceptable and should be retained. In contrast, rich-biased wealth-maximizing policies are acceptable only if their distributional consequences will actually be offset by additional taxes and transfers.Footnote ⁴⁴

But if current tax-and-transfer is insufficient and if offsets will not occur, there is no reason to distinguish between neutral and rich-biased policies. If policies are to be selected with redistribution in mind, we should also favor poor-biased policies over neutral policies, and very poor-biased policies over policies that are only mildly poor-biased. The designation of some policies as poor-biased and others as rich-biased should not affect our analysis at all; only the fundamental tradeoff between efficiency and redistribution should matter.

IV. Why Efficiency Is Not Biased: A More General Critique

Another way that efficiency might be rich-biased is in comparison to normative theories based on fairness or equity that produce results more favorable to the poor. Bias in this sense usually means bias in comparison to the speaker’s preferred alternative theory.

This formulation of bias is obviously problematic, since it begs the question by assuming the speaker’s preferred normative theory as the neutral baseline. The neutral baseline might be more convincing if there were some compelling and universally agreed-upon alternative to efficiency—but there is not. Philosophical theories that emphasize fairness over efficiency diverge from each other in a variety of ways. A critical legal theorist’s vision of fairness may favor the poor relative to efficiency analysis; but a minimal-state libertarian’s vision of fairness may favor the rich relative to efficiency analysis.

A more minimal formulation of this claim might be that Kaplow and Shavell’s elevation of efficiency is specifically rich-biased. Here, the alternative is more obvious. Kaplow and Shavell proposed that redistribution should not be conducted through legal policy; so the alternative is to redistribute through legal policy. But this formulation, too, merely buries the main disagreement. A core claim by Kaplow and Shavell is that focusing on efficiency through legal policy leaves everyone better off, including the poor. This might not be true for a variety of empirical reasons, but any rich-bias is not by design. Any claim of bias therefore must substantively engage with the assumptions of Kaplow and Shavell’s model, rather than broadly dismissing it as biased.

Again, allegations that efficiency is biased assume a neutral baseline that tends to hide assumptions rather than clarify them. Instead of focusing on whether a particular framework is or is not biased, we should ask whether it substantively fails or succeeds. We can consider the many important and well-known objections to the promiscuous invocation of efficiency in law, without throwing out efficiency analysis entirely.