Visual digital intermediaries and global climate communication: Is climate change still a distant problem on YouTube?

PLOS ONE Visual digital intermediaries and global climate communication top across contexts. Our results are consistent across data collections performed at different times, which allows us to make observations regarding the presence or absence of specific themes and actors in the search results. However, we do not use our data to support statis- tically significant statements such as the fact that the proportion of videos showing arctic themes for one country is higher or lower than for another. 4.2 Data collection The data for this study was collected from the YouTube Search API in May 2021. The collec- tion retrieved search results for all countries that can be specified in the API in their main official languages, based on videos published up until December 2020. We decided to use all languages that are listed as official for at least one of the country codes that can be specified in the YouTube API in order to be as complete as possible. We achieved satisfactory coverage of 232 countries and 85 languages, despite not being able to find translations for a few languages (which we document in the data released with the study). The accounts used to retrieve the data are set to not use personalisation. We used the following API settings: • We set order by ‘relevance’, the default search setting. • The search term or query (‘q’) was ‘climate change’ or its equivalent in each respective language. • Searching for all official languages meant that for some countries we made multiple distinct

queries. For example, in the case of Canada, we searched for both Canada/English/‘climate change’ and Canada/French/‘changement climatique’. • We used the RegionCode parameter to pass country information to the API. This parame- ter only retrieves videos that can be viewed (i.e. are not blocked) in the specified country. • We did not use the location and locationRadius parameters, which constrain the results to channels located in the specified radius, since they do not seem to be used in the default search settings. The data consists in the ranked results of 276 queries, which cover 232 countries and 85 languages. Our self-imposed limit was 400 results per search. We here only report find- ings based on those region/language pairs with 400 results. 253 queries returned the full 400 results, covering 219 countries and 62 languages. We focus the analysis particularly on top-20 rankings [48,49]. 4.3 Data preprocessing: Channel coding and image annotation To prepare the channel data, we coded all channels with videos in top-20 rankings by actor type, based on channel description. The coding scheme of 17 categories builds on previ- ous work [69,79], to include actors that are conventionally prominent or studied (e.g. legacy media, science, and civil society organisations), traditional actors that have conventionally been less prominent (e.g. large businesses, political parties, public agencies, religious organisa- tions), and actor types created for this context (e.g. user-led channels, online popular science

and education actors). The inter-coder reliability between two coders on a random sample of 10% of a larger data set of 1631 channels was Cohen’s kappa .81 (see supplementary mate- rial: coding schemes, channel categories). While the agreement was high in general, there was some disagreement on the code Popular science group, which we note here because this category became important in the analysis. This disagreement was analysed and resolved by discussion, as it had to do with miscommunication regarding the focus on non-institutional PLOS ONE https://doi.org/10.1371/journal.pone.0318338 April 8, 2025 8/ 23