World Happiness Report 2023 150 Averaging across genders. In chapter 4 of the World Happiness Report 2022 (WHR 2022), the authors95 report results from a study that assessed emotions, including happy/joy/positive affect, sadness, and fear/anxiety/scared over two years in the U.K. Prior work has found demographics like gender and age to impact patterns in language use more than personality and are thus important confounding variables to consider when analyzing language use.96 The authors in chapter 4 of the WHR 2022,97 separately derived (and then combined) gender-specific estimates from Twitter data using both Level 1 (LIWC) and Level 3 (contextualized word embeddings; RoBERTa) approaches.98 Twitter-estimated joy correlated at r = .55 [.27, .75] with YouGov reported happiness over eight months from November 2020 to June 2021. Gen 2 person-level aggregation – Summary Person-level Gen 2 methods are built on a decade of research using Gen 1 random feed aggregation methods based on the (in hindsight obvious) intuition that communities are groups of people who produce language rather than a random assortment of tweets. This intuition has several methodological advantages. First, person-level aggregation treats each person as a single observation, which can down-weight highly active accounts and minimize the influences of bots or organizations. Second, it paves the way for addressing selection biases as one can now weight each person in the sample according to their representativeness in the population. Furthermore, these methods can be applied to any digital data. Finally, these methods more closely reflect the methodological approaches in demography and public health that survey people and lay the foundation for tracking digital cohorts over time (Gen 3). Person-level aggregation can down-weight highly active accounts and minimize the influences of bots.
RkJQdWJsaXNoZXIy NzQwMjQ=