Introduction
In a series of influential works, Rizzo and Whitman (henceforth, RW) put forward several theoretical and practical challenges to behavioural paternalists’ attempts to design and implement welfare-enhancing paternalistic interventions in public policy (see, e.g., RW, Reference Rizzo and Whitman2009a, Reference Rizzo and Whitman2009b, Reference Rizzo and Whitman2018, Reference Rizzo and Whitman2020a and Reference Rizzo and Whitman2023; also WR, Reference Whitman and Rizzo2015 and Reference Whitman and Rizzo2021). RW’s works provide a systematic critique of behavioural paternalism’s conceptual foundations and of behavioural paternalists’ attempts to enhance individuals’ welfare. I lack the space here to engage with the many interesting themes examined in RW’s works. In this short paper, I focus on one foundational theme that figures centrally in such works, namely the theory of inclusive rationality (henceforth, IR) that RW put forward ‘as an alternative’ to the standard axiomatic conceptions of rationality (henceforth, SR) ‘shared by neoclassical and behavioral economics’ (WR, Reference Whitman and Rizzo2021, 382, italics added; also RW, Reference Rizzo and Whitman2018, 217). According to RW, SR is ‘excessively narrow [and] cannot capture the full depth and complexity of human choice’ (Reference Rizzo and Whitman2020a, xi and 433). For its part, IR ‘does not dictate the normative structure of preferences’ and encompasses a wide range of choice patterns that ‘do not fit nicely into the straitjacket of [SR]’ (ibid., 17 and 26).Footnote 1
The two main sections of this paper are structured as follows. The next section (‘Standard rationality versus inclusive rationality’) outlines the consistency conditions presupposed by SR and explicates the alleged contrast between RW’s theory of IR and SR. The third section (‘Inclusive rationality: a critical assessment’) articulates and defends a critical assessment of RW’s theory of IR. I shall argue for four main claims concerning such theory. First, the consistency conditions presupposed by SR can be defended against the main criticisms put forward by RW. Second, the proponents of SR can incorporate several insights provided by RW’s theory of IR without having to relinquish their reliance on SR’s consistency conditions. Third, RW’s theory of IR faces substantial falsifiability concerns, which seem more widespread and pervasive than those faced by SR. And fourth, RW’s theory of IR does not ground more informative and reliable evaluations of public policies’ welfare implications than SR. These four claims do not detract from the many merits of RW’s works. In particular, they do not bear against the main criticisms that RW articulate against behavioural paternalists’ attempts to design and implement welfare-enhancing paternalistic interventions in public policy.Footnote 2 In this perspective, my critical assessment of RW’s theory of IR can be seen as a constructive contribution to the ongoing discussion about RW’s theory of IR (see, e.g., Cowen and Dold, Reference Cowen and Dold2021, on a dedicated special issue) and the broader cross-disciplinary debate about rationality, welfare analyses and policy evaluation (see, e.g., Oliver, Reference Oliver2023, for recent contributions to such debate).
Standard rationality versus inclusive rationality
SR explicates the notion of rationality in terms of specific structural conditions on preferences. More specifically, to qualify as SR-rational, an agent’s preferences ‘must satisfy [the axioms of] completeness and transitivity, as well as certain corollaries [such as] independence of irrelevant alternatives’ (RW, Reference Rizzo and Whitman2020a, 16; also 80).Footnote 3 These axiomatic conditions constrain sequences of preferences, taken collectively, but place no substantive constraints on preferences, taken individually (see, e.g., Broome, Reference Broome, Frey and Morris1993, 52; Sugden, Reference Sugden1991, 760). As noted by RW, such axiomatic conditions ‘provide a logical foundation for […] the existence of utility functions [and make] economic models mathematically tractable’ (RW, Reference Rizzo and Whitman2020a, 53 and 81; also WR, Reference Whitman and Rizzo2015, 409 and 416). The idea, encapsulated in so-called representation theorems, is that if an agent’s preferences satisfy specific axiomatic requirements, then this agent’s choices can be represented as if the agent maximizes expected utility (see, e.g., Von Neumann and Morgenstern, Reference Von Neumann and Morgenstern1947, on situations of risk; Savage, Reference Savage1954, on situations of uncertainty). SR models do not aim to provide accurate characterizations of the neuro-psychological substrates of choices and are typically agnostic about such substrates (see, e.g., Fumagalli, Reference Fumagalli2013). In particular, the preferences figuring in representation theorems are commonly regarded as indexes of choices, but SR does not commit choice modellers to regarding preferences in general as reducible to or identical with choices (see, e.g., Beck, Reference Beck2024, on different conceptions of preferences).Footnote 4
RW’s theory of IR draws on interrelated descriptive, normative and prudential criticisms of SR. These criticisms can be explicated as follows. The consistency conditions presupposed by SR occasionally ‘provide a reasonable approximation of how people really behave’ (WR, Reference Whitman and Rizzo2021, 382). However, systematic descriptive violations of these conditions have been observed (see, e.g., RW, Reference Rizzo and Whitman2020a, ch. 1, WR, Reference Whitman and Rizzo2015). Moreover, an individual’s preferences may ‘violate the axioms of completeness and transitivity [without the individual being] irrational in any normatively significant sense’ (RW, Reference Rizzo and Whitman2020a, 75, italics added; also WR, Reference Whitman and Rizzo2015, 420). For one may be ‘discovering [or forming her] preferences […] during the process of choice’ (RW, Reference Rizzo and Whitman2020a, 58 and 81; also WR, Reference Whitman and Rizzo2015, 418). And in many cases, ‘the costs of completely rationalizing [one’s] preferences exceed the benefits of doing so’ (RW, Reference Rizzo and Whitman2020a, 81; also RW, Reference Rizzo and Whitman2018, 202). Therefore, an inclusively rational individual ‘will not, and should not, have complete and transitive preferences’ (WR, Reference Whitman and Rizzo2015, 419, italics added; also RW, Reference Rizzo and Whitman2020a, 239, WR, Reference Whitman and Rizzo2021, 383).Footnote 5
As to prudential considerations, RW hold that SR builds on consistency as the main welfare criterion and assumes that if individuals’ preferences violate completeness and transitivity, then such preferences fail to reliably track individuals’ welfare (see, e.g., RW, Reference Rizzo and Whitman2020a, ch. 6–7, WR, Reference Whitman and Rizzo2015). However, in their view, SR’s consistency conditions provide ‘no basis at all’ for determining which preferences track welfare (RW, Reference Rizzo and Whitman2020a, 18). For individuals ‘may have mutable preferences […] or no relevant preferences’ (ibid., 28). And individuals’ inconsistencies ‘can typically be resolved in more than one way’ (RW, Reference Rizzo and Whitman2023, 202; also WR, Reference Whitman and Rizzo2015, on the difficulties inherent in identifying welfare-optimal rates of saving and intertemporal discounting in specific choice settings). Moreover, abiding by SR’s prescriptions does not guarantee individuals to make welfare-optimal choices. For choices that violate SR’s axioms may be ‘adaptive to the circumstances [and] can increase the agent’s welfare’ (Rizzo, Reference Rizzo and Henderson2018, 193).Footnote 6
RW’s theory of IR draws on these interrelated descriptive, normative and prudential criticisms of SR to provide ‘an alternative’ to SR (WR, Reference Whitman and Rizzo2021, 382, italics added; also RW, Reference Rizzo and Whitman2018, 217). I shall expand on the specific tenets of RW’s theory of IR in the next section. For now, I note that contrary to SR, IR ‘does not dictate the normative structure of preferences’ and encompasses a wide range of choice patterns that ‘do not fit nicely into the straitjacket of [SR]’ (RW, Reference Rizzo and Whitman2020a, 17 and 26). In particular, IR allows that the set of ‘rational’ preferences may include preferences that are ‘inchoate, incomplete, inconsistent, mutable, and dependent on context’ (ibid., 26; also RW, Reference Rizzo and Whitman2018, 205).
Inclusive rationality: a critical assessment
In this section, I articulate and defend a critical assessment of RW’s theory of IR. I shall argue for four main claims concerning such theory, which respectively concern: the defensibility of the consistency conditions presupposed by SR against the main criticisms put forward by RW; the possibility of incorporating into SR several insights provided by RW’s theory of IR without having to relinquish SR’s reliance on consistency conditions; the falsifiability concerns faced by RW’s theory of IR; and the applicability of RW’s theory of IR to evaluating public policies’ welfare implications.
SR’s consistency conditions
RW correctly note that SR’s consistency conditions are violated in several choice settings (see, e.g., Gilboa et al., Reference Gilboa, Postlewaite and Schmeidler2009, on violations of completeness; Sugden, Reference Sugden1991, on violations of transitivity) and that various authors challenge the normative validity of such conditions (see, e.g., Aumann, Reference Aumann1962, on completeness; Anand, Reference Anand1993, on transitivity). However, SR’s consistency conditions can be defended against the main criticisms put forward by RW. Below I provide three replies to RW’s descriptive and normative criticisms.Footnote 7
First, the reported descriptive violations of SR’s consistency conditions tend to significantly decrease in presence of experienced decision makers (see, e.g., List, Reference List2003; Choi et al., Reference Choi, Kariv, Müller and Silverman2014) and in situations where individuals are given time and incentives to learn about the choice problems they face (see, e.g., Plott and Smith, Reference Plott and Smith2008; Oprea, Reference Oprea2020). Moreover, in recent decades the proponents of SR have developed several SR models which modify specific axioms so as to increase SR’s descriptive fit with the available empirical findings (see, e.g., Machina, Reference Machina, Durlauf and Blume2008, and Starmer, Reference Starmer2000, for reviews). In this respect, it would be of limited import to object that the models involving such modifications are more plausibly regarded as IR (rather than SR) models (see, e.g., RW, Reference Rizzo and Whitman2020a, 31, holding that ‘when behavioral economists invoke bounded rationality, they are in essence claiming that the bounds of [SR] models are not appropriate’). For despite modifying specific axioms, such models retain SR’s reliance on axiomatic consistency conditions on preferences. That is to say, RW are correct that early SR models (e.g., expected utility theory) face descriptive criticisms and that some SR models merely accommodate (rather than predict) individuals’ choices. Still, several modelling developments have occurred within SR over the last few decades, and various such developments are plausibly regarded as empirically progressive (see, e.g., Guala, Reference Guala2005, and Starmer, Reference Starmer2005, for illustrations).Footnote 8
Second, SR’s consistency conditions can be given a plausible normative defence. For instance, completeness is less demanding than what RW appear to presuppose, since it requires individuals to be able to specify their preferences over the alternatives figuring in the examined decision problems rather than over all possible alternatives (see, e.g., Grüne-Yanoff, Reference Grüne-Yanoff2021, 291; Gustafsson, Reference Gustafsson2022, sec. 3). And transitivity can be defended by pointing to the losses that individuals tend to incur by violating it (see, e.g., Grüne-Yanoff, Reference Grüne-Yanoff2021, 294–295; Gustafsson, Reference Gustafsson2022, sec. 4, on monetary and other welfare-relevant losses) and to individuals’ willingness to revise intransitive choices when they realize these choices’ intransitivity (see, e.g., Hands, Reference Hands2014, 401–402; Nielsen and Rehbeck, Reference Nielsen and Rehbeck2022, 2237–2239).Footnote 9 To be sure, the normative plausibility of completeness and transitivity may vary depending on what conception of preferences one presupposes (see, e.g., Mandler, Reference Mandler2005, 255–256, holding that completeness is more easily defended for individuals’ behavioural preferences than for preferences that encapsulate individuals’ judgements about their own welfare) and what SR models one examines (see, e.g., Sugden, Reference Sugden1991, 763, holding that the restrictions Savage’s theory imposes on what factors can be included in the description of a consequence hamper the defensibility of transitivity within such theory). Still, these dependencies do not generally bear against the normative plausibility of SR’s axioms. In this respect, it is telling that many of those who doubt the descriptive validity of SR’s axioms retain those axioms as a normative standard (see, e.g., Angner, Reference Angner2019, 203, on leading behavioural economists; RW, Reference Rizzo and Whitman2020a, 40, on leading behavioural paternalists).Footnote 10
And third, many reported descriptive and normative violations of SR’s axioms can be accommodated by modifying the descriptions of choice options presupposed by the purported counterexamples to such axioms (see, e.g., Broome, Reference Broome, Frey and Morris1993; also Fumagalli, Reference Fumagalli2020c, for critical discussion). By way of illustration, consider an individual who, faced with pairwise comparisons between food items A (apple), B (banana) and C (cake), exhibits the intransitive preference pattern A > B, B > C and C > A. One may accommodate this violation of transitivity by incorporating reference to what options are available to the individual into the description of each choice option. More specifically, let Ab indicate A when B is the other option available, Ba indicate B when A is the other option available, and so on. The individual’s preference pattern can then be re-described as Ab > Ba, Bc > Cb and Ca > Ac, which does not directly violate transitivity (see, e.g., Broome, Reference Broome, Frey and Morris1993, 54). To be sure, RW are correct that ‘allowing redescription whenever we encounter […] violations of the axioms’ would render SR’s axioms descriptively and/or normatively empty (Reference Rizzo and Whitman2020a, 70; also Anand, Reference Anand1993, 103; Bhattacharyya et al., Reference Bhattacharyya, Pattanaik and Xu2011, 146). However, the proffered re-descriptions of choice options are not equally plausible, and choice modellers are frequently able to demarcate whether specific factors (e.g., the price, spatial location and caloric content of specific food items) can be justifiably incorporated into the description of choice options by determining whether those factors do (or can plausibly) make a difference to individuals’ preferences between such options (see, e.g., Dreier, Reference Dreier1996, and Fumagalli, Reference Fumagalli2020c, for illustrations).Footnote 11
Compatibility of IR and SR
According to RW, the set of inclusively rational preferences may include preferences that are ‘inchoate, incomplete, inconsistent, mutable, and dependent on context’ (RW, Reference Rizzo and Whitman2020a, 26; also RW, Reference Rizzo and Whitman2018, 205). In their view, ‘a good case can be made for [the] inclusive rationality’ of several preference patterns that the proponents of SR regard as ‘biases’ (RW, Reference Rizzo and Whitman2020a, 17). However, SR appears to be more ‘inclusive’ than RW allege. For SR has the resources to accommodate a wide range of preference patterns that the proponents of IR deem to be incompatible with SR. In particular, the proponents of SR can incorporate several insights provided by RW’s theory of IR without having to relinquish their reliance on SR’s consistency conditions. To illustrate this, consider three putative inclusively rational ‘biases’ examined by RW, namely preference change, framing effects and self-regulation.Footnote 12
Preference change. Individuals’ preferences frequently change across time and choice settings (see, e.g., RW, Reference Rizzo and Whitman2020a, ch. 3). According to RW, preference change is incompatible with SR since SR models ‘typically [assume that] the agent has preferences that remain the same over time’ (ibid., 78). Moreover, RW hold, many instances of preference change are inclusively rational (ibid., ch. 3). However, the proponents of SR have developed several models of preference change (see, e.g., Dietrich and List, Reference Dietrich and List2011, and Strohmaier and Messerli, Reference Strohmaier and Messerli2024, for reviews) and are not committed to regarding preference change as irrational. To illustrate this, consider the so-called endowment effect. According to RW, the endowment effect ‘is not consistent with [SR] if switching costs are low and the value of the [involved goods] is small relative to the chooser’s wealth’ (Reference Rizzo and Whitman2020a, 13). Moreover, RW hold, many instances of the endowment effect are plausibly regarded as inclusively rational since ‘possession or ownership [of a good] may reflect important human values’ (ibid., 110). However, the proponents of SR may accommodate such instances of the endowment effect. For if possession or ownership of a good reflects individuals’ values, then SR allows choice modellers to include reference to these values (and those values’ influence on individuals’ preferences) into the description of choice problems (see, e.g., Fumagalli, Reference Fumagalli2020c). To be sure, one may point to various cases where choice modellers lack reliable epistemic access to individuals’ values (see, e.g., RW, Reference Rizzo and Whitman2020a, ch. 6–7). Yet, the existence of such cases does not selectively support IR over SR. For choice modellers frequently have reliable epistemic access to individuals’ values (see, e.g., Bradley, Reference Bradley2017, ch. 1). And in cases where choice modellers lack reliable epistemic access to individuals’ values, adopting IR does not per se yield more informative descriptive and/or normative insights about such values compared to SR.Footnote 13
Framing effects. Individuals’ preferences are frequently sensitive to framing effects (see, e.g., WR, Reference Whitman and Rizzo2015). According to RW, the sensitivity of individuals’ preferences to framing effects is incompatible with SR, but is often plausibly regarded as inclusively rational (see, e.g., RW, Reference Rizzo and Whitman2020a, ch. 1). However, the proponents of SR are not committed to regarding the sensitivity of individuals’ preferences to framing effects as irrational. To illustrate this, consider the case of defaults. As noted by RW (Reference Rizzo and Whitman2023, 206), several defaults reduce individuals’ decision-making costs and are regarded by individuals as recommendations from trusted sources. Still, RW’s claim that SR-rational individuals ‘would be affected by these clearly relevant factors’ (ibid., 206) does not bear against SR. For if some defaults reduce individuals’ decision-making costs and are regarded by individuals as recommendations from trusted sources, then SR allows that such defaults may rationally influence individuals’ decisions (see, e.g., Oliver, Reference Oliver2013). To be sure, one may point to various cases where choice modellers disagree as to whether specific defaults may rationally influence individuals’ decisions (see, e.g., RW, Reference Rizzo and Whitman2023, 206). Yet, the existence of such contested cases does not selectively support IR over SR. For several cases are not contested (see, e.g., Sunstein, Reference Sunstein2015, on various defaults concerning dietary and financial decisions). And in contested cases, adopting IR does not per se yield more informative descriptive and/or normative insights about the examined defaults compared to SR.
Self-regulation. Individuals frequently rely on self-regulation across a variety of choice settings (see, e.g., RW, Reference Rizzo and Whitman2009a). According to RW, individuals’ reliance on self-regulation is incompatible with SR, since SR-rational individuals ‘will simply select the best option from those available’ (Reference Rizzo and Whitman2020a, 3). Moreover, RW hold, individuals’ reliance on self-regulation is often plausibly regarded as inclusively rational (ibid., ch. 1). However, the proponents of SR have developed various models to accommodate individuals’ reliance on self-regulation (see, e.g., Gul and Pesendorfer, Reference Gul and Pesendorfer2001, on various models of temptation; Ross, Reference Ross2011, on various multiple-self models). Moreover, the proponents of SR are not committed to regarding individuals’ reliance on self-regulation as irrational. For nothing in SR excludes the possibility that, in a given decision problem, relying on self-regulation may be ‘the best’ option available to individuals. And although one may point to various cases where choice modellers disagree as to whether self-regulation is ‘the best’ option available to individuals (see, e.g., RW, Reference Rizzo and Whitman2009a), the existence of such contested cases does not selectively support IR over SR. For several cases are not contested (see, e.g., Fumagalli, Reference Fumagalli2024, on various cases involving harmful addiction). And in contested cases, adopting IR does not per se yield more informative descriptive and/or normative insights about the examined instances of self-regulation compared to SR.
Falsifiability concerns
RW hold that SR models ‘do sometimes pass falsification tests’, but often ‘do not perform […] well’ in terms of falsifiability (Reference Fumagalli2020a, 38). The idea is that choice modellers are frequently unable to assess the rationality of individuals’ choices since ‘rationality and irrationality are defined relative to subjective preferences that are typically unobserved and often unobservable’ (ibid., 408, italics added; also Dold and Rizzo, Reference Dold and Rizzo2024, 8–10). These remarks aptly emphasize the limited falsifiability of SR as an abstract mathematical framework (see, e.g., Blaug, Reference Blaug1992, ch. 4; Fumagalli, Reference Fumagalli2020b), but do not cast doubt on the falsifiability of specific SR hypotheses – i.e. hypotheses stating that particular individuals’ preferences satisfy SR’s axioms – compared to hypotheses based on RW’s theory of IR. For in primis, SR hypotheses are more amenable to empirical/experimental testing than the proponents of IR seem to presuppose. And second, RW’s theory of IR faces substantial falsifiability concerns, which seem more widespread and pervasive than those faced by SR. Let me expand on these two issues in turn.Footnote 14
The empirical implications of SR models are typically conditional on several auxiliary assumptions (see, e.g., Cubitt et al., Reference Cubitt, Starmer and Sugden2001, on assumptions concerning the adequacy of individuals’ incentives). Therefore, the hypothesis that particular individuals’ preferences satisfy SR’s axioms can rarely be tested independently of auxiliary assumptions (see, e.g., Bhattacharyya et al., Reference Bhattacharyya, Pattanaik and Xu2011, 142–143; Cubitt, Reference Cubitt2005, 208). In this context, the availability of some findings contrary to the empirical implications of SR models does not per se imply that reliance on SR’s axioms is unjustified. For in many cases, findings contrary to the empirical implications of SR models are more plausibly regarded as evidence against some of the auxiliary assumptions rather than evidence against SR’s axioms (see, e.g., Hausman, Reference Hausman1992, ch. 12). This, however, by no means implies that SR hypotheses are unfalsifiable. For choice modellers can test the validity of specific auxiliary assumptions by performing a series of experimental reproductions (see, e.g., Plott and Smith, Reference Plott and Smith2008). And these experimental reproductions may enable choice modellers to significantly reduce the set of factors that can be plausibly invoked to accommodate alleged violations of SR’s axioms (see, e.g., Fumagalli, Reference Fumagalli2016b, on the adequacy of individuals’ incentives). That is to say, if observed choices seem to contradict the hypothesis that individuals’ preferences satisfy SR’s axioms, then choice modellers should test auxiliary assumptions about ‘procedures, payoffs, context, instructions, etc. […] rather than conclude that [the involved individuals] are irrational’ (Smith, Reference Smith2003, 471). Yet, if the alleged violations of SR’s axioms persist once these auxiliary assumptions have been tested, then the hypothesis that individuals’ preferences fail to satisfy SR’s axioms is more plausible than the alleged failure of such auxiliary assumptions (see, e.g., Fumagalli, Reference Fumagalli2020c, 350; also Section ‘Standard rationality versus inclusive rationality’ on the reported violations of specific axioms and on SR models developed in response to such violations).Footnote 15
As to the falsifiability concerns faced by RW’s theory of IR, RW hold that IR does not ‘function exclusively as a normative concept [but also] as a positive research program […] for generating testable hypotheses’ (WR, Reference Whitman and Rizzo2021, 385, italics added). In their view, IR ‘incorporates many subsidiary questions with testable implications [about] whether (and how much) people learn over time [and] whether (and how) people adopt regimes of self-[regulation]’ (RW, Reference Rizzo and Whitman2020b, italics added). However, the proffered characterizations of the notion of IR are insufficiently specific to imply specific testable hypotheses about individuals’ learning and self-regulation (e.g., how much learning is implied by IR in particular contexts? Which instances of self-regulation are compatible with IR?). In this respect, it would be of limited import to object that RW’s theory of IR counsels choice modellers ‘to have some [epistemic] humility […] rather than indulging the impulse to find fault’ (RW, Reference Rizzo and Whitman2020c). For although such epistemic humility is commendable, addressing the falsifiability concerns faced by RW’s theory of IR would require the proponents of IR to provide clear and informative criteria for demarcating which choice patterns are incompatible with IR. Regrettably, the proponents of IR have hitherto failed to provide such criteria. In fact, even authors sympathetic to IR question the falsifiability of RW’s theory of IR (see, e.g., Rajagopalan, Reference Rajagopalan2021, 269, holding that RW ‘do not go far enough to explore […] difficult cases’ and calling the proponents of IR to identify clear cases of choice patterns incompatible with IR).Footnote 16
A proponent of IR may object that choice modellers ‘should not be guided exclusively by the falsifying goal of finding exceptions [and] should also engage in the confirming goal of finding more varieties of inclusive rationality’ (WR, Reference Whitman and Rizzo2021, 386, italics added). The idea would be that IR generates ‘useful and sometimes successful hypotheses’ (ibid., 385) and that ‘there are mountains of evidence for [IR consisting] in all manner of self-regulatory behaviors [and] learning over time’ (RW, Reference Rizzo and Whitman2020c, italics added). However, these claims appear to significantly overestimate the alleged empirical support for IR. For the proffered empirical evidence does not selectively support IR over SR (see, e.g., sub-sections SR’s consistency conditions and Compatibility of IR and SR). In this perspective, much purported empirical support for IR may be plausibly regarded as an artefact of the vagueness of IR and of the ensuing unclarity concerning the putative implications of IR.
Welfare analyses
According to RW, SR’s axioms are ‘analytical assumptions that are not welfare-relevant’ (Reference Rizzo and Whitman2020a, 17, italics added) and ‘consistency of choice [fails to provide] an adequate basis’ for welfare analyses (Rizzo, Reference Rizzo2024, 13, italics added). In their view, ‘there is no valid and convincing basis’ for determining which choices maximize welfare (RW, Reference Rizzo and Whitman2020a, 363). For although ‘in principle, we can objectively define the choices that will maximize health, or lifespan, [or] wealth’ (RW, Reference Rizzo and Whitman2020c), ‘the correct weighting of [choices’] benefits and costs is unavoidably subjective’ (RW, Reference Rizzo and Whitman2020a, 408, italics added). RW are correct that it is often difficult for choice modellers to identify which choices maximize welfare and that many proffered identifications of such choices are contested (see, e.g., Dold, Reference Dold2018). However, the existence of these contested cases does not per se license general scepticism about SR’s potential to ground informative and reliable evaluations of public policies’ welfare implications (see, e.g., Fumagalli, Reference Fumagalli2021). In this respect, it would be implausible to hold that SR’s axioms are ‘completely inadequate as a prescriptive standard’ (Rizzo, Reference Rizzo and Henderson2018, 193, italics added; also Berg and Gigerenzer, Reference Berg and Gigerenzer2010, 148, holding that ‘almost no empirical evidence exists documenting that individuals who deviate from [SR’s axioms] earn less money, live shorter lives, or are less happy’). For although abiding by SR’s axioms does not guarantee that individuals make welfare-optimal choices (see, e.g., Gilboa et al., Reference Gilboa, Postlewaite and Schmeidler2009, on cases where individuals’ consistent choices are based on inaccurate information about the available options), individuals who abide by SR’s axioms often tend to obtain higher welfare-relevant payoffs than they would obtain if they failed to abide by such axioms (see, e.g., sub-section SR’s consistency conditions on transitivity).
More generally, the point remains that RW’s theory of IR seemingly ‘lacks analytical clarity when it comes to concrete questions of [welfare] evaluation’ (Dold, Reference Dold2023, 6). And this lack of clarity, in turn, constrains this theory’s applicability to evaluating public policies’ welfare implications. To illustrate this, consider RW’s claim that within IR ‘the appropriate standard of well-being is the one you would impose on yourself’ (Reference Rizzo and Whitman2020c, italics added) and that ‘the desirability of acting [on SR’s axioms depends] on showing that failure to do so will result in bad consequences to decision-makers from their own point of view’ (RW, Reference Rizzo and Whitman2020a, 121, italics added). These remarks seemingly presuppose a radical subjectivist conception of welfare, according to which the extent to which an individual is well-off is a purely subjective matter, i.e. exclusively depends on the individual’s subjective judgements and attitudes towards her life. Yet, such conception of welfare is vulnerable to severe objections (see, e.g., Kagan, Reference Kagan2009; Parfit, Reference Parfit1984, appendix I) and does not ground informative and reliable evaluations of public policies’ welfare implications (see, e.g., Griffin, Reference Griffin1986, ch. 1–3, and Scanlon, Reference Scanlon1996, on various goods/experiences that can affect individuals’ welfare at least partly irrespective of individuals’ subjective judgements and attitudes towards their lives).
A proponent of IR may object that IR can ground informative and reliable evaluations of public policies’ welfare implications and that IR’s focus on subjective considerations enhances (rather than hampers) IR’s applicability to welfare analyses (see, e.g., Rizzo, Reference Rizzo2024, 13, holding that in their welfare analyses the proponents of SR ‘must admit that in back of choices are mental preferences’). The idea is that within IR ‘the ultimate standard by which individuals’ behavior is evaluated is the degree of successful attainment of goals in the actual environment in which they find themselves’ (RW, Reference Rizzo and Whitman2020a, 38, italics added). However, this evaluative standard does not per se ground informative and reliable evaluations of public policies’ welfare implications (e.g. how should choice modellers identify individuals’ goals? How is the degree of successful attainment of such goals measured? And are all individuals’ goals such that their attainment directly contributes to individuals’ welfare?). In fact, appealing to individuals’ ‘environment’ may further hamper IR’s applicability to evaluating public policies’ welfare implications. For the proponents of IR rarely provide precise specifications of which factors are plausibly taken to belong to individuals’ ‘environment’. And this paucity of precise specifications hampers choice modellers’ ability to ground informative and reliable evaluations of public policies’ welfare implications on appeals to individuals’ ‘environment’ (see, e.g., Hands, Reference Hands2014, 407, for similar remarks targeting the generic characterizations of individuals’ ‘environment’ presupposed by leading ecological rationality models).
A proponent of IR may further object that IR ‘could allow’ choice modellers to ground informative and reliable evaluations of public policies’ welfare implications ‘in a manner unrelated to the violation of consistency axioms […] by getting inside people’s heads as much as is feasible’ (WR, Reference Whitman and Rizzo2021, 385, italics added; also WR, Reference Rizzo and Whitman2018, 214, commenting on what ‘neuronal and behavioral responses to prediction errors […] we should expect on the part of [inclusively] rational actors’). However, the proponents of IR currently lack a suitable basis to ground informative and reliable evaluations of public policies’ welfare implications on empirical assumptions about neuro-psychological substrates. For many different (and often conflicting) models of the neuro-psychological substrates of choice have been advocated in the recent literature (see, e.g., Padoa-Schioppa and Schoenbaum, Reference Padoa-Schioppa and Schoenbaum2015). And despite the ongoing advances in neuro-psychological modelling, many prominent neuro-psychological models of choice are more plausibly regarded as ‘as-if’ models rather than accurate characterizations of the neuro-psychological substrates of individuals’ choices (see, e.g., Moscati, Reference Moscati2024, targeting leading ecological rationality models). Moreover, it is dubious that choice modellers’ evaluations of public policies’ welfare implications should be grounded on empirical assumptions about neuro-psychological substrates. For severe difficulties plague the proffered attempts to build neuro-psychological indexes of welfare (see, e.g., Fumagalli, Reference Fumagalli2019, on influential neuro-psychological indexes’ failure to track what many theories of welfare regard as individuals’ welfare). And prominent proponents of neuro-psychological indexes sharply disagree as to which indexes should be adopted to evaluate public policies’ welfare implications (see, e.g., Fumagalli, Reference Fumagalli2022, for illustrations).
Conclusion
In this paper, I have articulated and defended a critical assessment of RW’s theory of IR in light of the ongoing cross-disciplinary debate about rationality, welfare analyses and policy evaluation. The paper aimed to provide three main contributions to this debate. First, it explicated the relation between the consistency conditions presupposed by SR and the standards of rationality presupposed by RW’s theory of IR. Second, it provided a qualified defence of the consistency conditions presupposed by SR against the main criticisms put forward by RW. And third, it identified and discussed specific strengths and weaknesses of RW’s theory of IR in the context of welfare analyses and policy evaluation.
In their influential works, RW provide valuable critical insights concerning the descriptive/normative validity of SR and SR’s applicability to evaluating public policies’ welfare implications. However, as it stands, RW’s theory of IR is vulnerable to objections. These objections do not detract from the many merits of RW’s works and do not bear against the main criticisms that RW articulate against behavioural paternalists’ attempts to design and implement welfare-enhancing paternalistic interventions in public policy. Still, if correct, they challenge RW to qualify and/or better support their theory of IR.
As to future developments in the broader cross-disciplinary debate about rationality, welfare analyses and policy evaluation, three lines of research seem especially worthy of investigation, namely: assessing the prospects of a possible synthesis or partial convergence between SR and IR despite their several differences; further exploring the relation between SR, IR and other notions of rationality (e.g., ecological rationality) that figure prominently in the specialized cross-disciplinary literature; and probing the applicability of SR and IR to specific debates in welfare analyses and policy evaluation, as aptly showcased by RW’s influential works concerning paternalistic interventions.
Acknowledgements
I thank Malte Dold, Mario Rizzo and an anonymous reviewer for their comments on earlier versions of this paper. I also received helpful feedback from audiences at King’s College London, the University of Bolzano, the University of Pennsylvania, and New York University.
Funding statement
I acknowledge the support of the Foundations of the Market Economy Program, Department of Economics, New York University, for a brief research visit at New York University.
Open access
