Social ties and concern for global warming

Abstract

Recent research focusing on social factors affecting risk perceptions has suggested that social networks might help to explain why differences of opinion about climate change persist across segments of the lay public despite the scientific consensus. Even though concern for global warming in itself might seem irrelevant for most social ties, we show that it is significant enough to be reflected in the structure of social networks. To do this, we programmed a Facebook application that collected survey data on concerns and network data on friendships. We found that respondents tend to have friends with similar concerns as their own, the unconcerned respondents have fewer friends, and any two respondents who disagreed about the seriousness of global warming were less than half as likely to be friends. The results indicate that the structure of the social network may hinder changes in opinions, explaining why opinions persist despite the scientific consensus. The results suggest that the communication of climate science could be improved by strategies that aim to overcome these network effects.

Appendices

Appendix A: Local clustering coefficient

We did not observe the local clustering coefficient directly, but we could estimate it for those who have friends in the sample. Consider the full undirected Facebook network (N, G), where N is the set of nodes and $G \subset \{(i,j) \ \colon i,j\in N,\ i\neq j \}$ is the set of friendships.

Let N _i = {j ∈ N :(i, j) ∈ G} be node i’s set of friends and d _i the number of friends. The number of pairs of friends is p _i = d _i (d _i −1)/2. Let c _i = |{(j, k) ∈ G :j, k ∈ N _i }|/2 be the number of friendships between the members of N _i . Now the true local clustering coefficient is

$$t_{i}= \frac{c_{i}}{p_{i}}.$$

Let $\tilde {N} \subset N$ be the set of nodes we sampled. Let $\tilde {G} = \{(i,j) \in G \ \colon i\in \tilde {N} \text { or } j\in \tilde {N} \} \subset G$ be the set of friendships we sampled. Now, for each $i\in \tilde {N}$ we observed N _i .

Let $\tilde {d_{i}}= |N_{i} \cap \tilde {N} |$ be the number of i’s friends in the sample. We observed all pairs except the ones where both nodes were outside of the sample. The number of pairs we observed was then the number of all pairs less the unobserved pairs, which we denote by $\tilde {p}_{i} = d_{i}(d_{i}-1)/2 - (d_{i}-\tilde {d}_{i})(d_{i}-\tilde {d}_{i}-1)/2= d_{i}\tilde {d}_{i} - (\tilde {d}_{i}^{2} +\tilde {d}_{i})/2.$

Let $\tilde {c}_{i} = | \{ (j,k)\in G\ \colon j,k\in N_{i} \text { and } (j \in \tilde {N} \text { or } k \in \tilde {N}) \} |/2$ be the number of friendships between the members of N _i that we observed.

The estimated clustering coefficient is then

$$\tilde{t}_{i}= \frac{\tilde{c_{i}}}{\tilde{p_{i}}}. $$

This is what we could observe and use for our analysis.

Appendix B: Full regression tables

Here we report in full the regression tables of the main text. Tables 4 and 5 present all coefficient estimates of Tables 2 and 3, respectively.

Table 4 Regression analysis of number of friends and local clustering coefficient

Table 5 Logistic regression on the probability of friendship