Rizzo and Whitman’s Escaping Paternalism: Rationality, Behavioral Economics, and Public Policy — a review
It’s rare to find a book on behavioural economics that can pick at the flaws while presenting the best case for the other side. Mario J. Rizzo and Glen Whitman’s Escaping Paternalism: Rationality, Behavioral Economics, and Public Policy is not that rare book. While the book is excellent in many respects, Rizzo and Whitman rarely give the strongest argument for their behavioural economist foils.
That said, Rizzo and Whitman have provided possibly the most thorough and entertaining critique of the application of behavioural economics to public policy.
Their argument is multi-pronged. They disagree with the normative benchmark against which human decisions and behaviour are assessed. Alternative interpretations of the experimental results in the literatures on heuristics, biases and beliefs suggest that many of the supposedly irrational behaviours are superior in real-world contexts. Further, this literature is full of ambiguity and lacks generalisability, making it insufficient to guide policy. People are often aware of their constraints, so they can counteract or control them. People learn, particularly when there are material consequences to their decisions.
There is a knowledge problem in that policy makers don’t (and, in fact, can’t) know the preferences of those whose behaviour they seek to change. Instead, they fill in what they assume to be appropriate preferences. To top it off, paternalist policy makers are subject to the same biases as everyone else and face incentives that don’t necessarily align with citizen welfare. This hampers the development of good policy and sets us on the slippery slope to more restrictive paternalism.
I am sympathetic to all of those threads of argument, and Rizzo and Whitman provide a strong foundation for each. They cover a broad literature that a shallow dip into behavioural economics wouldn’t reveal (e.g. reading Daniel Kahneman’s Thinking, Fast and Slow or Dan Ariely’s Predictably Irrational). We often have a poor conception of why people act the way they do. People are heterogeneous. It’s hard to change behaviour in a complex system without unintended consequences. The knowledge problem is real. Supposedly coercion-free ideas (e.g. requiring opt-out retirement plans) often involve coercion (e.g. of employers). Their discussion in Chapter 10 of the features that can distinguish dangerous from innocuous interventions — self-imposed versus other-imposed, invited versus uninvited, informative versus manipulative — is excellent.
Rizzo and Whitman call their alternative benchmark “inclusive rationality”. Inclusive rationality is not conformance to a set of assumptions or rules of logic. Inclusive rationality includes experimenting with choices, discovering preferences and learning. It involves structuring your environment, possibly for purposes of self-control. While these behaviours don’t involve what Rizzo and Whitman call “puppet” rationality, they are sensible behaviours for real people.
They concede that people make mistakes, including by failing to act on their own preferences. However, I wish they had rigorously defined inclusive rationality and tested the boundaries. Throughout the book, it seems as though everything is inclusively rational. For example, they discuss action bias in the context of politicians liking to believe that something they did produced a good outcome. They write that “Even this mental process might be deemed broadly rational, once we allow for beliefs that have direct utility to the believer.” What evidence would persuade them that someone had made a mistake?
One question I kept asking throughout the book was: who exactly are these behavioural economists they are debunking? Rizzo and Whitman often state the behavioural economics canon on a topic, then cite papers that challenge it. But those challenges are usually by behavioural economists too! In many cases this is simply the typical passage of time — people propose ideas, future papers challenge and refine them.
For example, they describe quasi-hyperbolic discounting as the default option for analysing anomalies over time, then cite work by David Laibson — one of its main supporters — and other behavioural economists in rebuttal. Similarly, the evidence that auto-enrolment in retirement savings — the “iconic” case for behavioural paternalism — delivers less than it seems comes largely from Beshears, Choi, Laibson and Madrian, the group behind much of the early evidence on default effects. The debate about loss aversion is also largely (although not exclusively) one between behavioural economists. (At one point they concede that many of the claims they debunk come from “older and more popular” analyses. Many of their statements that “behavioural economists believe X” are better read as “Richard Thaler, Cass Sunstein or some other populariser stated X”.)
Many of these threads stop at a convenient place for their case against the “behavioural economist view”. They will cite the initial behavioural economics finding and the critical response, but often not the rejoinders.
Their examination of the Linda problem and the associated conjunction fallacy is a good example. After being given a description of Linda designed to be representative of a feminist, people are asked which is more probable: that Linda is a bank teller or that Linda is a bank teller who is active in the feminist movement. Many people choose the latter, even though a conjunction can’t be more probable than either of its parts. There are many good reasons to critique this experiment, as Gerd Gigerenzer and others have done. Maybe participants interpreted “more probable” in a way that differs from what formal logic requires. Rizzo and Whitman cover this well.
But there are experiments that deal with many of the objections. For instance, the same paper they cite for the Linda experiment also reports an experiment where people bet on a die with four green and two red faces. Participants could win $25 if their chosen sequence appeared anywhere in 20 rolls. With this real money at stake, two-thirds of participants chose to bet on GRGRRR over RGRRR. But every time the longer sequence shows up, the shorter one does too. The shorter one can also turn up on its own. No-one had to interpret the word “probable”. Rizzo and Whitman don’t mention this experiment (more on the debate here).
This pattern repeats through most sections. Rizzo and Whitman discuss Gal and Rucker’s argument against loss aversion, even citing a commentary in support, while ignoring the critical commentaries. They borrow Don Moore’s categories of overconfidence (although only in Chapter 8 when critiquing policy makers, not in Chapter 5 when making sweeping claims), without mentioning his findings on how stable the various forms of overconfidence are. They cite the replication crisis when it suits them. A footnote describes how the priming studies in Kahneman’s Thinking, Fast and Slow were grossly underpowered, yet their case for counteractive self-control uses studies that prime students to think about distractions. They rely on Roy Baumeister’s model of willpower as a resource that can be exhausted, despite the failed replications.
Presenting these debates does support their arguments urging caution in applying these ideas to policy. There is much uncertainty. But the evidential omissions leave a sour taste. It’s actually very Gigerenzerian — Gigerenzer has a habit of finding important weaknesses in experiments, then ignoring later experiments that debunk his debunking or show the limits of his interpretation (as I catalogued in one sequence of papers here).
Another omission is any examination of how behavioural economics is applied in practice. They know the Behavioural Insights Team (BIT) exists — they mention it in their public choice analysis of the issue and name David Halpern’s Inside the Nudge Unit as a leading work of behavioural paternalism — but don’t reference how BIT works or what types of interventions they roll out. BIT’s work generally involves observation phases to see what people actually do and how their behaviour varies. BIT recognise that there are surprises and unintended consequences, which is why they rely on field trials. It’s likely the most empirical form of policy you can find (alongside development economics). It also undercuts Rizzo and Whitman’s treatment of Cass Sunstein’s call for more randomised trials as mere rhetoric. The applied units are running those trials.
That is not to say there is nothing to criticise in BIT’s or other public policy-focussed nudge units’ way of working. There is. I have criticised them for many of the things Rizzo and Whitman identify. They can too quick to attribute behaviour to biases, assume what people want and substitute their own preferences when they know little about the people they want to influence. But from reading Rizzo and Whitman, you wouldn’t know that the practice on the ground is richer than the academic debates they have drawn from.
And what of the types of interventions implemented by applied units like BIT? Let’s take one of their most famous: social norms on tax payment letters. Stating that 9 out of 10 people pay on time increases on-time payment. Rizzo and Whitman might respond that this isn’t paternalism, as the letter serves the public purse, not the taxpayer. Fair enough. They might also argue this is inclusive rationality — there is information in the social norm! Maybe. Applied behavioural practitioners aren’t just interested in behaviour that can be classified as irrational. They’re interested in behaviour generally.
If you walk through the reams of BIT reports, you’ll see things like personalised reminder messages, commitment devices for job seekers and friction removal. Rizzo and Whitman question whether reminders are paternalistic or even count as nudges. They seem to think a nudge must target a “bias”, having earlier ruled out limited attention and memory as biases. Are these objectionable under Rizzo and Whitman’s framework? When job seekers make their own plans, they use the kind of self-imposed commitment Rizzo and Whitman praise in Chapter 6. Would they object to a job centre offering people that option? In Chapter 10 they concede that acceptable nudges might exist, but they don’t seem to do much looking.
The discussion of default enrolment in retirement savings plans shows how one-sided their selection of case studies can be. Auto-enrolment has been suggested by behavioural economists as a way to overcome present bias in people’s retirement planning. I have sympathy for their critique — I believe Australia’s mandatory retirement savings should be abolished — but they skip a stronger test of their own framework: Richard Thaler and Shlomo Benartzi’s Save More Tomorrow. In its original form, people can opt in to a plan that funnels a fraction of any future pay rise into retirement savings. It addresses loss aversion, in that take-home pay never goes down, and present bias, in that the contribution increases take effect in the future. Once in, status quo bias keeps them there. It is exactly the kind of self-imposed commitment device that Rizzo and Whitman champion. They know the paper exists — it sits in the reference list and they have written about it before (although normally in combination with auto-enrolment) — but they never discuss it in the text. What do Rizzo and Whitman make of this scheme in its original form?
The result is a good book burdened by an uneven treatment of the material. It could have been great — and many parts are — but read it for the general argument rather than as a well-balanced critique of behavioural economics in public policy.