An increasing number of economists analyze subjective welfare data – which records a subject’s “happiness” or “life satisfaction” – as a complement to more traditional money-based measures of wellbeing such as income or consumption. Both the promise and the pitfalls of subjective welfare (SW) measures have been widely discussed, including in this blog here and here and here. One major challenge is the concern that fixed personal characteristics (such as someone’s “natural optimism”) determine SW responses to a far larger degree than time-varying economic factors. If that is the case then the usefulness of SW data for informing economic policy is not clear. Now two recent papers teach us more about the interpretive difficulties of SW in the presence of fixed individual characteristics.
The concern that fixed characteristics influence SW is encapsulated in the “set point theory” where, in this view, equilibrium levels of subjective welfare are determined by temperament or personality traits that are largely fixed by adulthood. If this is the case, then gains in income arising through, say, a salary raise or a lottery win may increase SW but only temporarily before a return to baseline levels. The implication of this concern is a limited scope for economic growth or economic policies/programs to sustainably affect these welfare measures. The relative intransigence of subjective welfare with respect to macro-economic shocks has been noted in several papers, including this piece by Angus Deaton.
The set point theory implies that we live on a “hedonic treadmill” and that our SW is largely unresponsive to changes in material conditions. One natural question, if we accept that most of us live life on the hedonic treadmill, is whether really poor individuals who face frequent challenges to material survival also live on the treadmill. Perhaps among the most vulnerable or poor the treadmill hasn’t started yet and there may be large gains in happiness from income improvements. A recent working paper from Galiani, Gertler, and Undurraga directly explores this question by investigating the effects of the provision of high quality housing on subjective welfare among residents in extremely poor communities in El Salvador, Uruguay, and Mexico.
Under this program, basic pre-fabricated houses were supplied by a Latin American NGO and, due to limited supply, allocated to residents in targeted neighborhoods through lottery. A previous companion paper establishes that these new houses are a significant upgrade across many dimensions of housing quality. So these appear to be desirable homes that improve many aspects of material life on a long-run basis.
In each selected community, the houses were offered to lottery participants in two phases 8 months apart. However end line data was collected at the same point in time for all households regardless of whether it received a house in Phase 1, Phase 2, or not at all.
It turns out that one standard SW measure, a binary measure of overall life satisfaction, is 40% higher for the Phase II winners, when compared with the non-winners. Phase II winners have had a new house for 16 months. For Phase 1 winners, who have had a house for 24 months, life satisfaction is only 16% higher than non-winners. This difference suggests that 60% of the SW gain from free housing that was seen at 16 months after house construction goes away in the following 8 months. It appears that even for low-income households in disadvantaged communities, the hedonic treadmill kicks in for large positive gains like a new house.
(One inferential challenge to this conclusion is the possibility that SW gains from the receipt of housing, or the winning of a lottery, are partly due to mimetic effects, and that winners in the second phase value housing more in part because they were phase 1 lottery losers and thus value a phase 2 lottery win all the more. While it may be a stretch – I have no reason to prefer this explanation over the alternative forwarded by the authors – it is also consistent with results. UPDATE [Nov. 6th, 2015] Subequent correspondence with the authors has indicated that this alternative explanation is far less likely. Please see the addendum at the bottom of the blog post.)
Another recently published paper, by Caliendo, Cobb-Clark, and Uhlendorf explores how subjective beliefs around the efficacy of job search affect the likelihood of finding a job. I discuss this paper here because (a) subjective beliefs of this type are related to SW assessments, and (b) they appear to affect employment, which is related to many traditional economic measures that policy cares about.
The particular beliefs studied refer to the degree of control people have in determining life success – if someone believes that much of what happens in life is beyond control of the individual then that person exhibits what is termed an external locus of control. On the other hand, individuals with an internal locus of control believe that much of what happens in life is within their control. It has been shown in psychological studies that individuals with an internal locus of control are more likely to set higher goals, persevere in difficult settings, and achieve successful outcomes.
The locus of control (loc) when measured through survey typically asks about various perceptions on control in the life course and on whether successes and failures experienced in life are deserved. Individuals with an internal loc tend to be male and younger and have more years of schooling. In addition they tend to have somewhat different personality traits in so far as they are more open, conscientious, and extraverted.
The authors analyze repeated interviews of unemployed workers in Germany. As the initial interview was conducted two months after the start of the unemployment spell and then continued at regular intervals, it is not possible for unemployment duration to affect the loc perceptions.
The headline finding is that unemployed individuals with greater internal control more actively search for jobs. Individuals who score high in internal control measures and low in external measures submit 2.23 more job applications than their opposite counterparts. The authors find similar results when internal and external loc individuals are propensity matched on the basis of numerous observables including work history and education – it is important to control for these observables as they may capture productive factors such as cognitive ability that are correlated with both locus of control and job search strategies.
It appears that what links the locus of control to job search efforts is the belief about payoffs to the job search – individuals with an internal loc believe that the efforts they expend in a job search will be more effective. And this belief does seem to be reflected in results – individuals with an internal loc have a 13% higher monthly probability of exiting unemployment with a job.
The unifying point in these new papers is that there are characteristics of the individual, presumably mostly fixed by adulthood, that determine many subjective measures of welfare as well as affect economic behavior. We are still left with the question of how useful SW is for understanding the intersection of human welfare with economic policy if indeed much of the variation in SW measures are determined by fixed factors beyond the control of most available policy levers.
Addendum [Added November 6th, 2015]: In brackets above, I put forward a possible alternative interpretation to the findings in Galiani, Gertler, and Undurraga. It now seems that this alternative is far less plausible. The authors have clarified in correspondence to me that the Phase 1 and Phase 2 lotteries are in separate communities and it is unlikely that Phase 2 communities were aware of Phase 1 lotteries. Here I quote direct correspondence from Raimundo Undurraga:
"We assume that it’s quite unlikely since the Phase 1 lotteries were conducted in other communities than Phase 2 lotteries. Furthermore:
i) Communities are located in a sparse geographical territory, which makes unlikely that they know each other. The 74 communities in our sample are distributed in 3 countries, and within each country, in multiple states and municipalities. For example, in the case of El Salvador, our study includes 23 communities that are distributed in 6 different states throughout the country; and,
ii) This is not a government program. The program is implemented by an NGO in a limited number of slums, which limits the popularity of the intervention."
Raimundo, thanks very much for the correction and for your further thoughts.
Join the Conversation