World Happiness Report 2023 157 measurement of well-being in multiple languages based on limited training data to “fine-tune” these models. Beyond measurement, research is also needed on how social media is used differently across cultures. For example, research indicates that individuals tend to generate content on social media that is in accordance with the ideal affect of their culture.132 We are beginning to see the use of social media- based indicators in policy contexts. Foremost among them, the Mexican Instituto Nacional de Estadística y Geografía (INEGI) has shown tremendous leadership in developing Twitter-based well-being measurements for Mexican regions. Well-being across cultures. Beyond cross-cultural differences in social media use, as the field is considering a generation of measurement instruments beyond self-report, it is essential to carefully reconsider the assumptions inherent in the choice of measured well-being constructs. Cultures differ in how well-being—or the good life more generally—is understood and conceptualized.133 One of the potential advantages of language-based measurement of the good life is that many aspects of it can be measured through fine-tuned language models. In principle, language can measure harmony, justice, a sense of equality, and other aspects that cultures around the world value. Ethical considerations The analysis of social media data requires careful handling of privacy concerns. Key considerations include maintaining the confidentiality and privacy of individuals, which generally involves de-identifying and removing sensitive information automatically. This work is overseen and approved by institutional review boards (IRBs). When data collection at the individual level is part of the study design – for example, when collecting language data from a sample of social media users who have taken a survey to train a language model – obtaining IRB-approved informed consent from these study participants is always required. While a comprehensive discussion on all relevant ethical considerations is beyond the scope of this chapter, we encourage the reader to consult reviews of ethical considerations.134 Conclusion and outlook The approaches for assessing well-being from social media language are maturing: Methods to aggregate and sample social media data have become increasingly sophisticated as they have evolved from the analysis of random feeds (Gen 1) to the analyses of demographically-characterized samples of users (Gen 2) to digital cohort studies (Gen 3). Language analysis approaches have become more accurate at representing and summarizing the extent to which language captures well-being constructs – from counting lists of dictionary keywords (Level 1) to relying on robust language associations learned from the data (Level 2) to the new generation of large language models that consider words within contexts (Level 3). The potential for global measurement. Together, these advances have resulted in both increased measurement accuracy and the potential for more advanced quasi-experimental research designs. As always with big data methods – “data is king” – the more social media data that is being collected and analyzed, the more accurate and fine-grained these estimates can be. After a decade of the field developing methodological foundations, the vast majority of which are open-source and in the public domain, it is our hope that more research groups and institutions use these methods to develop well-being indicators around the world, especially in languages other than English, drawing on additional kinds of social media, and outside of the US. It is through such a joint effort that social-media-based estimation of well-being may mature into a cost-effective, accurate, and robust complement to traditional indicators of well-being. It is our hope that more research groups and institutions use these methods to develop well-being indicators around the world.
RkJQdWJsaXNoZXIy NzQwMjQ=