A large interdisciplinary literature on the relationship between age & subjective well-being (happiness) has produced very mixed evidence; these authors argue that this is due to potential sources of bias

Kratz, Fabian, and Josef Brüderl. 2021. “The Age Trajectory of Happiness.” PsyArXiv. April 18. doi:10.31234/osf.io/d8f2z

Abstract: A large interdisciplinary literature on the relationship between age and subjective well-being (happiness) has produced very mixed evidence. In this paper we argue that this is due to potential sources of bias that may distort the assessment of the age-happiness relationship. Most biases tend to produce a spuriously U-shaped age trajectory. In contrast, applying our suggested specification to German panel data we find a (nearly monotonic) declining age happiness trajectory.

Fig 4b Predicted age-happiness trajectories 

Summary and Conclusions

How aging affects happiness is an important research question for the social and behavioral sciences. Our literature review demonstrates that many conflicting age trajectories have been reported in the literature. As this state of research is quite unsettling for the science of happiness, we discuss—informed by recent advances in the methodology of causal analysis—model specifications used by researchers in this field. Altogether, we identify four main biases that may distort the age trajectory of happiness. By using the German SOEP data, we show that distortions may be huge producing even qualitatively different conclusions. We demonstrate that by using different combinations of mis-specifications it is possible to generate (almost) every trajectory that has been reported in the literature. With a model specification that avoids these four biases, we find an age-happiness trajectory that declines slowly over adulthood (altogether about half a scale point). The decline comes to a halt and we observe even a small increase (about one tenth of a scale point) during the golden ages. Afterwards, in old age a very steep decline in happiness sets in.

From these results we derive several conclusions that pertain to future research on happiness.  The overarching conclusion is that SWB scholars should take causal reasoning seriously in their future research. They should precisely define their research question and explicitly justify their model specification chosen according to the research question (these conclusions do not pertain to SWB research alone, but to all kind of social research as Lundberg et al. (forthcoming) argue forcefully).

Qualifying the research question before estimating age-SWB profiles is essential. Is the main research aim to describe how happy the living population is, or how SWB develops with rising age? If the aim is to answer the second research question about aging and thus to estimate a causal effect of age on SWB, scholars should not use (repeated) cross-sectional data, because these may be affected by mortality selection bias. And there is no cure for this with only cross-sectional data available. Only with panel data following the same respondents over time mortality selection bias can be fixed.

Even when using panel data, scholars must carefully consider potential sources of under^Band overcontrol bias and select an estimation approach that strictly relies on within-person variation to minimize mortality selection bias. In our empirical illustration with the SOEP, the less familiar sources of bias, (i.e., overcontrol and mortality selection bias) cause more severe distortions than does undercontrol bias. We illustrated that selective mortality exhibits drastic consequences on the association between age and subjective well-being affecting even qualitative conclusions: Mortality selection systematically removes the unhappiest of the oldest 21 old and therefore every approach that relies somehow on between persons variation underestimated the deteriorating impact of aging on subjective well-being (especially among the oldest old).

Avoiding these misspecifications is not only important for future research on the age trajectory of happiness but also for any kind of happiness research. Age is usually (and well justified) used as a control variable when investigating other determinants of happiness. Using mis-specified age trajectories can severely bias estimates of such treatment effects: the bias in the control variable age transfers to the treatment effect of interest (a formal statement of this so-called “bias transfer” can be found in Ranjbar & Sperlich, 2019). Therefore, it is important to use flexible parametrizations of the age effect in happiness research more generally.  Including linear and quadratic age terms only might be problematic.