We used a tool called @Botometer in our just-released paper to estimate the prevalence of automated Twitter accounts in the online discussion of #climate change, around the time Trump announced the U.S. withdrawal from the #ParisAgreement. Short thread. https://www.tandfonline.com/doi/abs/10.1080/14693062.2020.1870098

1. We did this study in 2018 and 2019, when we purchased and analyzed the pretty vast data. @Botometer may have been critiqued, but is really still the industry standard. It's what we used. “Botometer is state-of-the-art in Twitter bot detection research.“ https://www.researchgate.net/profile/Stefan_Suetterlin/publication/334139727_Team_learning_in_cybersecurity_exercises/links/5d1a241e299bf1547c8eec06/Team-learning-in-cybersecurity-exercises.pdf#page=29

2. Second, we are not saying definitively that each tweet and account in our list are automated. Rather, these are estimates, these are most likely to be bots, and behave most like automated accounts. We uniformly used a probability threshold, based on the literature.

3. Third, in our comparison of different discourses/topics found in our Structural Topic Model, we applied the same threshold, and found that some were more likely to be above the “likely bot” threshold. Those were also more likely to be skeptical of climate science and policy.

4. Our data is from four years ago, and bot detection software we used is closer to that time, and so this is a shifting problem. This study took a long time, and had to apply @Botometer over a period of time.

5. Our main bot detection work was during fall, 2019. Most of the critique papers are from 2020. This is how science works: you apply tools available, they get critiqued, better tools are developed, those get used. /End