In defense of behavioral welfare economics

ABSTRACT

In The Community of Advantage, Robert Sugden advocates an opportunity-oriented framework for normative analysis, positions it a substitute for behavioral welfare economics, and criticizes the latter. This paper distills the logic underpinning the main approaches to behavioral welfare economics, addresses Sugden's criticisms, and identifies some limitations of his alternative approach.

KEYWORDS:

• Behavioral welfare economics
• Bernheim–Rangel approach
• opportunity criterion
• welfare economics
• The Community of Advantage

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Readers familiar with Bernheim and Taubinsky (2018) will notice a change of terminology. There I referred to judgement critiques as fallibility critiques, and lumped coherence critiques and implementation critiques together under the heading of consistency critiques. I have since found the original terms to be a source of some confusion because poor implementation is a form of fallibility, and because it obscures the important distinction between coherence critiques and implementation critiques. It is also worth emphasizing that my classification of critiques does not include broader philosophical objections to desire-theoretic accounts of well-being; see, for example, Parfit (1984), Kagan (1998), and Hausman (2012).

2 One such argument invokes justifications for self-determination in the tradition of classical liberalism: my views about my life are paramount because it is, after all, my life. A second entails the Cartesian principle that subjective experience is inherently private and not directly observable, which renders each of us uniquely qualified to assess our own well-being. Modern libertarian philosophers such as Nozick (1974) describe self-determination as a fundamental right rather than a means to an end, and construe that principle as constraining the legitimate scope of government. This position is not easily reconciled with conventional welfare analysis.

3 Consider as an example an individual who exhibits ‘weakness of will’. Some would say that this weakness results in optimization failure because it prevents the individual from following their true preferences (Dworkin, 1971; New, 1999). Yet someone who behaves this way is still exercising will and expressing preferences. Thus, references to ‘weakness of will’ typically reflect disagreements about proper objectives rather than problems with optimization. The fact that these disagreements may be intrapersonal (in other words, that preferences may differ from one moment in time to another) raises other issues, but does not change the fact that critiques involving ‘weakness of will’ implicitly attempt to impugn direct judgments.

4 There is an ordinal difference between $U(\cdot \mid f)$ and $V\left(\cdot \right)$ for frame f if there is some x and y such that $V\left(x\right)>V\left(y\right)$ and $U(x\mid f)\le U(y\mid f)$.

5 Formally, a set of unbiased choices does not have a ‘true preference’ representation if it violates the Weak Axiom of Revealed Preference (WARP). Indeed, if the consumption set is finite and the unbiased choice correspondence is defined for all conceivable menus, WARP is necessary and sufficient for the existence of a ‘true preference’ representation. (This conclusion follows from the classic result of Arrow, 1959.) For an infinite set, additional technical conditions are required.

6 If W is cyclic, then it may be the case that it deems every available option ‘improvable’, in which case there is no best option.

7 Technically, the result assumes that, for every menu X, there is some frame f for which the decision problem (X,f) lies within the WRD, and that we observe C on the entire WRD. This requirement is less demanding than one might think. Obviously, we never actually observe all of these choices. Instead, we use a smaller set of observed choices to parameterize models that imply choices on the complete domain. For example, once we estimate a model with biased beliefs, we can use the model to impute bias-free decisions for all opportunity sets.

8 In The Community of Advantage (Sugden, 2018, p. 57), Sugden writes that this instruction is ‘rather unhelpful’. On the contrary, in practical applications, our framework can produce bounds on welfare effects that acknowledge the ambiguity implied by context-dependent choice while nevertheless providing useful policy guidance; see Bernheim et al. (2015) for an empirical example. In our personal correspondence, Sugden expresses regret for including this remark, ‘especially as p. 166 of [The Community of Advantage] accepts the idea that cost-benefit analysis might use imprecise individual preferences to arrive at imprecise measures of cost and benefit’ (Sugden, 2020).

9 Present focus implies a particular form of context dependence: choices differ according to whether they are ‘in the moment’ or ‘in advance’. We can think of naivete as involving a second dimension of context dependence. The following thought experiment helps to clarify this point. Imagine constructing an automaton that perfectly mimics the consumer's choice rule at all points in time and under all contingencies. In one setting (naturalistic choice), the consumer makes a choice in the current period knowing that she will also make all future selections. In a second setting (transparent choice), we provide her with the automaton's decision rule and ask her to make a selection in the current period, knowing that the automaton will make all future selections. A naive consumer will make different selections in these two settings even though they are objectively equivalent. The naturalistic choices suffer from characterization error, but the transparent choices do not. While we do not ordinarily have the opportunity to observe these transparent choices, we can infer them from the $\text{β}\delta$ model.

10 While Ambuehl et al. (2020, august) do not examine context-dependence within the set of simply framed choices, the method does not assume that these choices are context-independent.

11 As explained in Bernheim and Taubinsky (2018), relative transparency versus opacity can create a presumption concerning characterization failure, which one can diagnose by examining data on comprehension. Contrary to Sugden's claim, there are other routes to identifying instances of characterization failure, such as marshalling direct evidence on elements of decision processes, including memory, attention, and learning. We discuss these other approaches in Bernheim and Taubinsky (2018). For an application involving neurobiological evidence on hedonic forecasting, see Bernheim and Rangel (2004).

12 Because unincentivized responses have their limitations, it is worth noting that the analyst can also deploy incentivized approaches. For example, when the decision is complexly framed, the subject may be willing to pay a positive amount to learn the terminal payment, but when the decision is simply framed, she may be unwilling to pay anything to learn the principal amount and rate of interest.

13 In contrast, many dozens of successful empirical applications explicitly or implicitly apply the Bernheim-Rangel approach (see Bernheim & Taubinsky, 2018).

14 While Moral Realists would disagree, such disagreement is arguably itself a matter of opinion.

Additional information

Notes on contributors

B. Douglas Bernheim

B. Douglas Bernheim is the Edward Ames Edmonds Professor of Economics in the Department of Economics at Stanford University. He is also a Research Associate of the National Bureau of Economic Research, and a Senior Fellow of the Stanford Institute for Economic Policy Research.