News Media Coverage of Climate Change in India 1997–2016: Using Automated Content Analysis to Assess Themes and Topics

ABSTRACT

News media play an important role for public awareness and perception of climate change – and thus citizens’ behavior. Few studies focus on media coverage in poor and developing countries such as India – the third-largest polluter and an important player in global climate change policies. Further, even these few studies on Indian media coverage span short time periods, focus on specific events, and evaluate pre-defined themes. Applying LDA topic modeling on 18,224 climate change articles published between 1997 and 2016 in two Indian newspapers, we find that climate change coverage in India has increased substantially in the last 20 years. We categorized the coverage into 28 different topics related to four overarching themes: “Climate Change Impacts”, “Climate Science”, “Climate Politics”, and “Climate Change and Society”. Climate change has gained more media attention since 2007 in general with a particular increase in focus on the theme “Climate Change Impacts”. Implications about shifting media discourses and its potential to educate people and change policies are discussed.

KEYWORDS:

• Climate change
• media coverage
• India
• topic modeling
• longitudinal analysis

Acknowledgements

We thank the two reviewers for their time and effort to improve our manuscript. Additionally, we are grateful to Lea Hellmüller, Uma Shankar Pandey, and Martin Wettstein for their feedback on previous versions of this manuscript. We also thank Anna Staender, Katherine M. Engelke, Korinna Olivia Lindemann and Rafael Schwab for their external coding of topics.

Disclosure statement

No potential conflict of interest was reported by the authors.

ORCID

Tobias R. Keller http://orcid.org/0000-0001-5263-4812

Valerie Hase http://orcid.org/0000-0001-6656-4894

Jagadish Thaker http://orcid.org/0000-0003-4589-7512

Daniela Mahl http://orcid.org/0000-0002-5330-6885

Mike S. Schäfer http://orcid.org/0000-0002-0847-7503

Notes

1 In the following, we differentiate between themes and topics: overarching themes, e.g. “Climate Politics”, consist of several, more specific topics, e.g. International summits and Regional summits.

2 Articles were prepared for LDA topic modeling by lower case conversion, removal of numbers, punctuation, white space and stop words as well as stemming (Feinerer et al., 2008; Grimmer & Stewart, 2013; Günther & Quandt, 2015). As, based on Zipf's law, we can expect many of the words in our corpus to occur only as rarely as once (Li, 1992), we decided to trim the corpus by removing words that occurred less than five times.

3 Statistical fit of a model was indicated though perplexity of models across k. Perplexity describes model fit when training the models on training data (75% of the corpus) and evaluating them on held-out data (25% of the corpus) (Blei et al., 2003; Jacobi et al., 2015). Robustness was indicated through topics being stable across different k. Interpretability was indicated through coherence and exclusivity of topics, judged based on top terms of each topic across different k (Quinn et al., 2010).

4 We defined ten topics to be of the theme “Background” (see appendix, Table B). These do not mainly refer to climate change but instead list public events in India, describe awards, school competitions, etc. Though internally consistent, these background topics are not further analyzed here. They are regarded as background noise irrelevant to our analysis concerning the coverage of climate change (Jacobi et al., 2015; Maier et al., 2018) given that when selecting articles containing our keyword as seldom as once, we also pick up articles in other contexts which mention these terms only in passing.

5 This topic is one of the three topics that show low validity and robustness and should hence be considered with caution (for details see Table A in the appendix). This topic specifically merges with Forests and Animals in further robustness tests and is therefore not clearly distinguishable. We marked the three topics that scored low both on validity and robustness measures with an asterisk.