Comments
Inequality, well-being, and the problem of the unknown reporting function
I expect some Forum readers will be interested in a couple of recent PNAS papers that discuss new causal evidence on the link between income and wellbeing.
I recommend reading both papers in full (they are short and open-access). In this link post, I provide a few quotes from Kaiser & Oswald's commentary to highlight the importance and limitations of the Dwyer & Dunn findings.
Disclaimer: Ryan Dwyer is my colleague at the Happier Lives Institute and Caspar Kaiser is one of our trustees but neither paper was produced by HLI.
Every politician, in every nation and in every era of history, eventually has to face a complex and emotive question. Should I try to redistribute money from my richer citizens to my poorer citizens? If so, by how much? This is a timeless issue. The appropriate answer to the question turns crucially on a claim that goes back hundreds of years to, for example, the philosopher Jeremy Bentham: “All inequality is a source of evil—the inferior loses more in the account of happiness than by the superior is gained.” (1) In an ideal world, a hypothesis of this sort would be tested in a giant randomized controlled trial (RCT), perhaps funded by a body such as the National Science Foundation of the United States. However, no funding body is likely to provide the necessary millions of dollars to run that experiment-until now. In a remarkable and important contribution to conceptual science and practical public policy, Ryan Dwyer and Elizabeth Dunn (2) have—with the help of millionaire donors—run an RCT that comes close to that ideal.
Dwyer and Dunn create an experiment in which assignment to treatment is random. The authors’ work is an example of a more general movement in modern social science in which earlier correlational research is checked with experimental and quasiexperimental designs (e.g., ref. 3). They add also to an emerging causal literature on cash transfers and well-being in low- and middle-income countries (summarized in, e.g., ref. 4). Dwyer and Dunn deliberately field the same kind of cash transfer in several different economic contexts. In this way, they connect two kinds of literature and demonstrate that cash transfers do indeed improve recipients’ self-reported well-being across a wide variety of settings. There are large causal effects that persist over at least a 6-mo period.
One complication—formally recognized more than half a century ago (6), but it continues to be hotly debated in political life—is that tax-funded redistribution may distort incentives and thereby dampen economic growth. If so, redistribution could act to raise well-being via a more equal income distribution but at the same time could lower well-being by decreasing the size of the ‘pie’. Assessing the relative importance of these countervailing forces—as discussed in conventional economics courses—remains a priority.
Loss aversion is a second complication. Income losses are known to loom larger than gains. In the short term, therefore, the well-being losses of those who are taxed may exceed the gains of net recipients. Based on earlier work (7), Dwyer and Dunn suggest that voluntarily giving money away can increase people’s well-being. Whether that is also true in the case of forced redistribution is an open question. It may eventually be possible to answer that by building on current studies of loss aversion (such as refs. 8 and 9).
It clearly seems as though Dwyer and Dunn give causal evidence for a curved well-being shape. Yet, as they mention in passing, although they do not elaborate, their paper depends on an untested assumption. It is that the relationship between reported well-being and the underlying actual well-being is linear. Whilst scattered work hints at possible linearity (10–13), nobody currently knows whether that is true. It depends on how humans use language when they answer happiness kinds of questions. This is a particular version of a generalized difficulty outlined in a recent piece by Bond and Lang (14) and previously discussed in Oswald (12). The earlier literature on psychophysics also grappled with this (15). Without knowledge of the ‘reporting function’, we cannot be sure that we can treat well-being data as ratio-scale measurements, which is required for statements like “three times more happiness”.
We would like to emphasize that our instinct is that the authors’ finding is correct and that Dwyer and Dunn have written a superb paper of lasting importance. Further research in this area will still be appropriate. It may include qualitative work, perhaps with in-depth interviews on respondents’ scale-use, as well as quantitative work that would systematically map subjective responses to observable cardinal quantities (as in refs. 12 and 15). Currently, the reporting-function problem is fundamental, little recognized, and so far unsolved.
