Climate change discourse in U.S. history textbooks from California and Texas

Abstract

Anthropogenic climate change is a scientific fact, but U.S. public discourse around the issue remains mired in controversy, including in education. Our study leverages natural language processing methods to give a precise look into the extent to which climate change-related topics are covered in 30 of the most widely used high school history textbooks in California and Texas. We find that history textbooks situate climate change-related topics within narratives of U.S. progress and development, and focus on the role of government in climate action. Consistent with analyses of science curricula, we also find that history textbooks emphasize controversy in climate discussions. Despite differences in state-level standards, the content of textbooks in California and Texas is surprisingly similar in the extent and nature of climate change-related discourse. Our study indicates that history textbook reform is an important arena for expanding and improving climate change education.

Keywords:

• California
• climate change
• history
• Texas
• textbooks
• scientific consensus

Acknowledgements

We extend special thanks to Emma Dolan for research assistance with earlier versions of this paper. Our work also benefited from research support provided by Sebastian Andrews and Li Lucy. Members of the Computational Sociology Workshop and Comparative Sociology Workshop at Stanford University contributed valuable comments on earlier drafts.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 We wanted a portrait of the climate education most students were intended to receive at the time the study was conducted. Because we did not sample by publication year, we also do not systematically examine how the content of these textbooks changes over time. Although publication dates of the textbooks vary slightly, the books share in common their wide usage at the time they were sampled.

2 A token is defined as a sequence of characters grouped together as a useful semantic unit for processing, usually between spaces or punctuation (Manning, Raghavan, and Schütze 2008)

3 Because topic loading measures tend to be right-skewed within each topic, we define a highly associated sentence within a given topic as having a topic loading measure that is two standard deviations above the corpus mean.

4 This frequency estimate excludes common articles, conjunctions, and other terms without content (“stop words”) such as “the,” “and,” “as,” etc.

5 We removed three topics out of an original 27 topics because they were driven by large portions of text related to wartime weaponry involving gasoline, as well as wartime rationing of gasoline, rather than discussing gas and oil as related to climate or environmental issues. An additional two topics were removed because they discussed social controversies unrelated to climate change or environmental issues. We opted to leave terms that pulled these sentences in our key term list despite their dual usage because of their relevance to overall energy and pollution discussions. The omitted topics are discussed in Appendix C.

6 Robustness checks for the topic modeling results were performed in multiple ways. First, we performed the topic modeling separately by state, to ensure that no individual state dominated the across-corpus themes discussed above. When topics are generated by state individually, 20 out of 22 climate-relevant topics are comparable between California and Texas, and the higher-order themes are consistent between the two. We examined output from an alternative LDA topic modeling package in Python that leverages the MALLET toolkit (McCallum 2002). Finally, we tested an alternative topic modeling algorithm, Structural Topic Modeling (STM; Roberts et al. 2013), which allows the incorporation of a state-identifying covariate to differentiate between state corpora. For both the alternative LDA specification and STM specification, higher-order themes are consistent when examining the resulting topic output.

Additional information

Funding

This work was supported by the Stanford Institute for Human-Centered Artificial Intelligence (HAI) and Stanford’s Center on Philanthropy and Civil Society (PACS).

Notes on contributors

Hannah K. D’Apice

Hannah K. D’ Apice is a PhD candidate studying International and Comparative Education at Stanford University’s Graduate School of Education. Her work examines the development and diffusion of global norms around education, as well as the institutional conditions that enable marginalized groups to have greater visibility and leadership in education. Prior to her doctoral studies, she managed multi-state randomized controlled trials evaluating K-12 education curricula and programming, as well as worked as a teacher in Texas and Singapore. She has an M.A. in Sociology and an M.A. in International Education Policy Analysis, both from Stanford University, as well as a B.A. in Political Science from Columbia University.

Patricia Bromley

Patricia Bromley is an Associate Professor of Education and Sustainability at Stanford University and Co-Director of the Center on Philanthropy and Civil Society (PACS). At PACS she directs the Global Civil Society and Sustainable Development Lab. She teaches courses related to the sociology of education, nonprofit organizations, and global education policy in the International and Comparative Education Program. Her research spans a range of fields including organization and management theory, comparative education, and the sociology of education, covering the substantive topics of sustainable development, minority and human rights, nonprofits/civil society, and globalization.