Asking for Directions: Using Classification and Regression Trees to Identify the Pathway to High Life Satisfaction

Background: Life satisfaction is a cognitive self-appraisal of one's current circumstances, influenced by both external circumstances ("bottom-up" factors) and features of personality ("top-down" factors). Life satisfaction is consistently associated with physical and mental health outcomes, making it a public health priority; however, life satisfaction in the US has been precipitously decreasing. The purpose of this study was to identify factors that could be addressed by U.S. policy for the purpose of enhancing population-level life satisfaction.Methods: Data for this study came from Wave 1 of the Global Flourishing Study. Seventeen actionable/modifiable variables were selected as covariates. They included a wide variety of behavioral, socioeconomic, psychological, and health factors. All variables were dichotomized for the Classification and Regression Trees (CART) analysis, a decision-tree approach used for problems of classification. This supervised, nonparametric, multivariate method was employed to identify the most accurate set of classifiers for the continuous outcome: Life satisfaction. CART offers several advantages over traditional ordinary least squares regression, including making fewer statistical assumptions, automatically detecting nonlinear relationships, and handling a large number of variables without the risk of multicollinearity. The CART model was trained and tested using a 70/30 split of the US sample (n=38,312). The model was further tested for replication in a combined international sample of similar Western countries (United Kingdom, n=5,368; Australia, n=3,844).Results: Of the 17 potential covariates, only 5 were represented in the models: balance, satisfaction with relationships, purpose, mental health, and comfortability with income. Each of the three models had similar numbers of splits (5, 7, 5). In all three models, the CART analysis identified life balance as the primary node. Those with low balance (the left-hand split) had much lower life satisfaction than those who self-reported balance (nearly a 3-point difference in life satisfaction, on average). For those with low balance, mental health was the second node for all datasets, and relationship satisfaction was the final split in the test and international samples. For those with good life balance in the US (the right-hand split), the tree structure was slightly more variable, as splits on purpose, relationships, mental health and comfortability with income were observed. For the international sample, splits only occurred on relationships and purpose.Conclusions: The results clearly show the primary factors related to life satisfaction. Life balance was the most important, followed by mental health and satisfying relationships for those with low life balance. For those with high life balance, relationships, purpose, and comfortability with income (in the US only) were most important. These findings partially diverge from our hypothesis, as balance served as the most important correlate of life balance. The study findings not only reveal the most important correlates of life satisfaction, replicated across multiple test samples, but also those that were surprisingly less important (e.g., physical health, participation in a social or religious group, exercise). The findings have significant implications for interventions and policy development since all factors included were chosen are modifiable.