Men deny more than they believe about climate change on Twitter (X)

Data and methods I have used twitter data available on Kaggle [37]. It is claimed to be one of the most extensive dataset available on Twitter that relates to human attitudes and climate change. With over 13 years of temporal coverage, over 15 million tweets geographically dispersed worldwide. Each tweet is analyzed across seven informational dimensions: geolocation, user gender, aggression, climate change attitude and sentiment, deviations from historical temperatures, topic model- ing, and data on environmental disaster occurrences. The data used by [37] suggests that these dimensions are the result of testing and analyzing numerous cutting-edge supervised and unsupervised machine learning algorithms and techniques. The previous studies have heavily relied on text analysis and descriptive techniques of coun- trywide variation of sentiments and tweets about climate change and have produced useful information. These studies have used text analysis approaches to uncover the pattern and vari- ations of climate change discussion across the globe. In order to have significant insights across the vital opposite groups (believer and denier) of climate change, a robust analytical model is required. Given the three stances, suitable model choices are ordered logit or multinomial. Respecting the opinions of individuals, it will be highly appropriate to not to rank the groups i.e. to make any order of three groups. In this case, it is useful to go for the multinomial model that does not consider any order between the outcome categories. Therefore, in this way, add-

ing to the contributions discussed in the literature review section, I will use multinomial logis- tic regression model to assess the relationship between the dependent variable (stances: believer, neutral or denier) and the key independent variables. Where the independent vari- ables are temperature, gas emission, impact of overconsumption, tweet language (aggres- siveness) and gender. These variables are given in the dataset and are categorized based on the expression of individuals on the twitter. Thus, these categorized expressions will help under- standing the likelihood of individuals taking their stance (believer, or denier, or neutral) for the climate change. While all the aforementioned independent variables are important, for the purpose of this paper, I have maintained a focus on gender during detailed interpretation of result. To assess the relationship between the qualitative outcome variable (stances of climate change by individuals via tweets), I have used the multinomial logistic regression represented by the following equation. lnðpj=pjÞ ¼ Aj þ BjX The logistic model is a generalization of the binary outcome of standard logistic regression comparing each category of the outcome (climate believer and denier in this case) to a referent category (neutral). The equation assumes, there are J total categories of the outcome, indexed by the subscript j, and the number of comparisons is then J– 1. The equation is written in terms of the logit of the outcome, which is a comparison of a particular category to the referent

category, both denoted as πj in the equation. The j subscript on both the intercept, Ai, and slope, Bj, indicate that there is an intercept and a slope for the comparison of each category which is a generalized case of binary logistic model. The predicted probability for each predictor can be computed by generalizing the above equation for standard logistic, using the following equation. pj ¼ 1=ð1 þ X eAjþBjX Þ The model is used by earlier studies on climate change in the context of agriculture [38] and is the best choice for this data as the outcome has three unordered categories- believer, neutral and denier and there are sufficient numbers of cases across the categories over such a PLOS ONE Men deny climate change on Twitter PLOS ONE | https://doi.org/10.1371/journal.pone.0303007 February 13, 2025 4 / 9