Using Absolutist Word Frequency from Online Searches to Measure Population Mental Health Dynamics

Abstract

The assessment of population mental health relies on survey data from representative samples, which come with considerable costs. Drawing on research which established that absolutist words (e.g. never) are semantic markers for depression, we propose a new measure of population mental health based on the frequency of absolutist words in online search data (Absolute Thinking Index; ATI). Our aims were to first validate the ATI, and to use it to model public mental health dynamics in France and the UK during the current COVID-19 pandemic. To do so, we extracted time series for a validated dictionary of 19 absolutist words, from which the ATI was computed (weekly averages, 2019-2020, n = 208). We then tested the relationship between ATI and longitudinal survey data of population mental health in the UK and France. ATI was linked with survey depression scores in the UK, r = .68, 95%CI[.34,.86], β = .23, 95%CI[.09,.37] in France and displayed similar trends. We finally assessed the pandemic’s impact on ATI using Bayesian structural time-series models. These revealed that the pandemic increased ATI by 3.2%, 95%CI[2.1,4.2] in France and 3.7%, 95%CI[2.9,4.4] in the UK. Mixed-effects models showed that ATI was related to COVID-19 new deaths in both countries β = .14, 95%CI[.14,.21]. Our results demonstrate the validity of the ATI as a measure of population mental health (depression) in France and the UK. We propose that researchers use it as cost-effective public mental health “thermometer” for applied and research purposes.

A Lost Idea in Psychology: Observation as Starting Point for the Scientific Investigation of Human Behavior

“…However, direct observation is sometimes impossible (e.g., behaviors mediated by technology, hidden by individuals). Another option we can suggest, more adapted to behaviors that are not directly observable, is the analysis of the traces of these real behaviors found in official statistics or big data (Adam-Troian et al, 2022;Harlow & Oswald, 2016;Peters et al, 2021). A third option, the most indirect one discussed here, consists in conducting descriptive exploratory studies, preliminary surveys, interviews, or collecting testimonies (e.g., Dezecache et al, 2021).…”