| 英文摘要: | Anthropogenic climate change represents a global threat to human well-being1, 2, 3 and ecosystem functioning4. Yet despite its importance for science and policy, our understanding of the causes of widespread uncertainty and doubt found among the general public remains limited. The political and social processes driving such doubt and uncertainty are difficult to rigorously analyse, and research has tended to focus on the individual-level, rather than the larger institutions and social networks that produce and disseminate contrarian information. This study presents a new approach by using network science to uncover the institutional and corporate structure of the climate change counter-movement, and machine-learning text analysis to show its influence in the news media and bureaucratic politics. The data include a new social network of all known organizations and individuals promoting contrarian viewpoints, as well as the entirety of all written and verbal texts about climate change from 1993–2013 from every organization, three major news outlets, all US presidents, and every occurrence on the floor of the US Congress. Using network and computational text analysis, I find that the organizational power within the contrarian network, and the magnitude of semantic similarity, are both predicted by ties to elite corporate benefactors. More attention needs to be given to the intersection of the natural sciences, social sciences, and the private sector, to uncover the structural roots of why, in the face of overwhelming scientific consensus, only 44% of Americans believe that anthropogenic climate change is happening, and only 14% are ‘very’ worried about its consequences5. We understand relatively little about these questions because, with a few exceptions6, 7, 8, 9, most popular and scholarly attempts to explain widespread doubt about climate change have focused on individual-level factors using attitudinal surveys or psychological experiments10, 11, 12, 13, 14, 15. Although important, this research has focused on outcomes rather than causes—and on individual attributes within the public rather than the institutional actors who produce contrarian information, the social network within which they are embedded, and the flows of resources that underwrite it. Very little research has been able to take this broader approach and explore these essential aspects because it is often difficult for researchers to obtain the necessary data using conventional methods. Emerging methods in computational social science16, 17, 18, 19 make possible a new approach whereby it is possible to comprehensively examine both the network structure and discursive influence of the climate change counter-movement at a large scale. The data used here include two interrelated parts: the first being the full institutional and social network structure of climate change contrarianism, and the second being its complete collection of written and verbal texts. This comprehensive social network is made up of 4,556 individuals with ties to 164 organizations involved in promulgating contrarian views. The individuals in this bipartite network include interlocking board members, as well as many more informal and overlapping social, political, economic and scientific ties. The organizations include a complex network of think tanks, foundations, public relations firms, trade associations, and ad hoc groups. I explain in detail in the Supplementary Methods how I constructed this network, its representativeness, and the variables I collect on each organization. A central, and empirically unanswered, question concerns the extent to which the private sector influences the production and diffusion of contrarian information. Research has suggested, but not yet empirically tested, that ExxonMobil (EM) and the Koch family foundations (KFFs) may have played a particularly important role as corporate benefactors7, 8. I therefore obtained Internal Revenue Service data recording whether or not any of the organizations in the contrarian network received funding from these corporate actors between 1993–2013. The KFFs are the philanthropic arm of Koch Industries, and include the Charles G. Koch Charitable Foundation, the Claude R. Lambe Charitable Foundation, and the David H. Koch Charitable Foundation. It is important to note that although many of these organizations receive corporate funding from a wide variety of sources, these two corporate actors supply the most reliable and theoretically important across-time indicators of corporate involvement. Although the media and politicians often echo contrarian discourse emphasizing ‘debate’ and ‘controversy’9, 20, we know relatively little about the magnitude of such influence and the potential covariates that might explain it. I apply advancements in computational text analysis to measure the external semantic influence of these organizations. To do this I collected the entirety of all written and verbal texts about ‘climate change’ or ‘global warming’ from 1993–2013 from every contrarian organization (40,785 documents containing over 39 million words). These texts include the totality of available press releases, published papers, website articles, scholarly research products, policy studies, and conference transcripts (see Supplementary Methods for more information). To examine which contrarian organizations’ ideas were successfully achieving influence in media and politics, I use latent semantic analysis21 to compare information in contrarian texts to information in all verbal and written texts about climate change between 1993 and 2013 from three major news outlets (14,943 documents), all US presidents (1,930 documents), and every occurrence on the floor of the US Congress (7,786 documents). The media texts come from LexisNexis and include the left-leaning New York Times, the right-leaning Washington Times, and the centrist USA Today. The written and verbal presidential texts come from The American Presidency Project, and the US Congress texts were obtained from the United States Government Printing Office. I use a common form of latent semantic similarity analysis that uses singular value decomposition, and calculates cosine similarity scores between two texts, which is expressed mathematically as:
where A is one text from an organization and B is one text from news media or politics. The cosine similarity of A and B ranges from a score of 0 (no similarity) to a score of 1 (perfect similarity). To prepare the texts for similarity analysis, I transformed them into a document term matrix, which includes stripping all whitespace, stemming using the Porter algorithm, converting all words to lower case, and removing all English stop words, sparse terms, numbers, and punctuation. I then use the lsa package in R to calculate cosine similarity coefficients for every organization by comparing their individual texts to the texts in the same year in news media, presidential, and Congressional data. I computed these for all texts and all years between 1993 and 2013. I then aggregated the mean of the coefficients by year, to assess if contrarian discourse as a whole became more similar over time to the discursive fields they intended to influence. Last, I use multivariate regression to predict organizational differences in these semantic similarity scores. This approach has several unique theoretical and empirical advantages. First, moving beyond the tendency to focus on individual-level attitudes about climate change, the network approach taken in this study captures the broader social structural arrangements in which contrarian information is actually produced. Second, given the discursive nature of climate change politics, the textual focus on writing and speech is an ideal way to investigate the issue. Collecting the total population of texts in the contrarian network sidesteps biases inherent in small sample sizes that hamper previous studies. Last, these data are naturally occurring phenomenon, and thus avoid many of the pitfalls that nag survey research, psychological experiments, or qualitative interviews. In Fig. 1 I present a bipartite graph of the global structure of the climate contrarian network, illustrating all ties between individuals (small black nodes) and organizations (large red nodes). Organizations create and exchange information in this network through these individuals, both formally and informally, at climate change conferences, board meetings, media strategy workshops, and political action committee gatherings. Descriptively, I find that the global structure of this network is such that there are not multiple components, and instead there is a more densely connected region, which is flanked by more loosely connected individuals and organizations.
| http://www.nature.com/nclimate/journal/v6/n4/full/nclimate2875.html
|
