ables 2 and 3 presents summary statistics for the continuous and categorical variables,respectively. The data set covers more than 4,000 individuals. It is an unbalanced panel with different time series support per individual. At best, there are 48 consecutive time waves per individual (on average about 18 observations per individual). Compared to other panel studies on SWB, the time series support of this data set is relatively long. The response rates for the variables of interest are very high. Thus, I do not expect significant bias due to item non-response. The survey designers intended a representative sample of US individuals. People were asked to participate and did not self-select themselves. This mitigates worries about endogenous sample selection. Nevertheless, a small endogenous component remains,since people still accept to be part of the sample. But clearly, this is a concern in any type of survey data and equally applies to every experimental study. The potential attrition bias is less severe for individuals than for firms. If firms leave a panel due to bankruptcy, they basically “die”. If people leave the panel, they are likely still alive and simply stop participating.If this non-participation is unrelated to the unobserved component, and not confined to specificsub-populations, this is no problem.