Gaussian Process-based statistical causal relationships in the 2022 climate Twittersphere

An interactive app for an exploratory case study on statistical causality relationships in the climate change-related Twitter debate during 2022.

Using the daily Tweets in 2022 that were pertinent to climate change [5], a text-based sentiment time-series signal was constructed using the methodology of [3,4]. We subsequently assessed, using the Gaussian Process-based framework of [1,2], the statistical causal relationships between the daily sentiment signal and the signal of the daily number of Tweets from two user communities (pro-climate and climate change denialists, [5]). The existence of causal relationships was investigated in the mean, as well as the mean and covariance of the Gaussian Process, using time-series lags of 1,3 and 5 days when fitting the GP model.

Technical references:

1. Zaremba AB, Peters GW. Statistical Causality for Multivariate Nonlinear Time Series via Gaussian Process Models. Methodology and Computing in Applied Probability. 2022;24(4):2587-632.
2. Zaremba AB. Assessing causality in financial time series. UCL (University College London); 2022.
3. Chalkiadakis, I., Yan, H., Peters, G.W. and Shevchenko, P.V., 2021. Infection rate models for COVID-19: Model risk and public health news sentiment exposure adjustments. PLoS One, 16(6), p.e0253381.
4. Chalkiadakis IM. Statistical natural language processing and sentiment analysis with time-series: embeddings, modelling and applications. Heriot-Watt University, School of Engineering and Physical Sciences; 2022.

Context and data reference:

5. David Chavalarias, Paul Bouchaud, Victor Chomel, Maziyar Panahi. The new fronts of denialism and climate skepticism: Two years of Twitter exchanges under the macroscope. 2023. ⟨hal-04103183v2⟩