World Happiness Report 2023 136 However, because life satisfaction is generally more widely surveyed than affective well-being, five of the six studies used life satisfaction as an outcome against which the language-based (affect) estimates were validated; only one study23 also included independent positive and negative affect measures to compare the language measures against (at the county level, from Gallup). Thus, taken together, there is a divergence in this nascent literature on geographic well-being estimation between the predominant measurement methods that foreground affective well-being (such as sentiment systems) and available data sources for geographic validation that often rely on evaluative well-being. This mismatch between the well-being construct of measurement and validation is somewhat alleviated by the fact that–particularly under geographic aggregation– affective and evaluative well-being inter-correlate moderately to highly. As we will discuss in this chapter, recent methodological advancements have resulted in high convergent validity also for social-media-predicted evaluative well-being (e.g., see Fig. 5.5: : Life Satisfaction Model). If social media data is first aggregated to the person-level (before geographic aggregation) and a language model is specifically trained to derive life satisfaction, the estimates show higher convergent validity with survey- reported life satisfaction than with survey-reported affect (happiness). Thus, specific well-being components should ideally be measured with tailored language models, which can be done based on separately collected training data.24 Figure. 5.1 showcases international examples in which different well-being components were predicted through Twitter language, including a “PERMA” well-being map for Spain estimating levels of Positive Emotions, Engagement, Relationships, Meaning, and Accomplishment,25 a sentiment-based map for Mexico,26 and a life-satisfaction map for the U.S.27 Photo by Junior Reis on Unsplash
RkJQdWJsaXNoZXIy NzQwMjQ=