Fig. 6: Differences in expression between US-based academics and the general US Twitter population.

Comparison of the temporal dynamics of academic expression with the general US Twitter population from January 2016 to December 2022. Data comprise n = 7,724 US academics and n = 61,259 general US Twitter users. US-based academics were sampled from a larger, politically active balanced sample of 26,555 academics (from an overall balanced sample of 99,274) by ensuring that they tweeted in English and had at least one politically oriented tweet in both the early (January–June 2016) and late (July–December 2022) periods. Panels show distributions, smoothed trends and variability across key behavioural and political metrics. For the stance measure: climate action, techno-optimism, behavioural adjustment, cultural liberalism and economic collectivism, the natural scale ranges from −1 to +1. In contrast, the other measures: toxicity (bounded between 0 and 1), emotionality/reasoning (a ratio typically close to 1) and egocentrism (ranging from 0 upward), are plotted on their inherent scales. These scale choices preserve the natural variation in each metric. Each data point represents the monthly average for each group, smoothed via LOESS with 95% confidence intervals around the trend line (shaded regions).