This is fundamentally a descriptive statistics question (how do X and Y go together) rather than a causal inference question. Descriptive claims still have value; it is clear that both self-assessed happiness and use of language have omitted causes not explored in this research. If anything, this is mostly useful as an exercise to critique operationalization of emotion as sentiment scores in NLP.