Money, Methodology, and Happiness: Using Big Data to Study Causal Relationships between Income and Well-being

How is people’s happiness determined by economic factors such as their income? Big data are essential to answering this question, but there is disagreement about the amount of evidence for causal relationships that different types of analysis give. This chapter reviews different approaches to analysis. First, it is argued that most existing literature both underclaims regarding the evidence for causality given by some types of analysis of big data, such as correlational analyses, and overclaims for other types of analysis, such as those involving panel data. Thus, even correlational data can be informative to the extent that associations are generally rare and that theoretical targets and alternatives are fully specified and given prior probabilities. Second, a new methodological problem is identified for a specific model of the income-rank relationship. According to the income rank hypothesis, people’s well-being is determined not by their income but by the ranked position that their income within a social comparison group. It is shown by simulation that spurious rank effects can occur in regression analyses if there is noise in measured income, but that this problem can be reduced with the use of robust regression techniques. A new analysis of a large dataset, the Panel Study of Income Dynamics, is reported. The results show that income rank effects are not reduced by the use of robust regression techniques, suggesting that previous support for the income rank hypothesis is not due to a noise-related artefact.