Seismic Risk: The Biases of Earthquake Media Coverage

The capacity of individuals to cope with threatening situations depends directly on their capacity to anticipate what will come next. The media should play a key role in that respect, but an extensive analysis of earthquake media coverage by the international news reveals systematic biases. Exploring a corpus of 320 888 news articles published by 32 worldwide newspapers in 2015 in English, Spanish or French, we found that the press covers a very small number of events: 71% of the news about seismic events was dedicated to only 3 earthquakes (among the 1559 of magnitude 5+). A combination of frequency and content analysis reveals a typical framing of the ‘earthquake news’. Except for the ‘Nepal quake’, the duration of the coverage is usually very short. The news thus tends to focus on short-term issues: the event magnitude, tsunami alerts, human losses, material damage, and rescue operations. Longer-term issues linked to the recovery, restoration, reconstruction, mitigation and prevention are barely addressed. Preventive safety measures are almost never mentioned. The news on impacts show a peculiar appetency for death counts, material damage estimates and sensationalism. News on the response tends to emphasize the role played by the international community in helping the ‘poor and vulnerable’. The scientific content of the coverage is often restricted to mentions of

Despite the increasing influence of social media, newspapers remain major gatekeepers in 72 the process of selection and dissemination of the news proposed by press agencies to national 73 and local audiences (Harcup & O'Neill, 2017). For risk managers, they remain an important 74 tool for reaching a wide range of people (Cottle, 2014). To this respect, one can expect the press 75 not only to inform, but also to provide the public with the knowledge to reduce disaster risks of playing the role of "crisis catalyst" (Boin et al., 2008). Comparing the news treatment of a 95 real earthquake with that of a false quake prediction, Smith (1996) concludes that "the interest 96 in drama at the expense of public affairs interferes with good scientific reporting." In general, 97 scientists denounce the tendency of the press to search for "culprits" and "accountability" and 98 for "stirring up old rivalry and exaggerating conflicts" (Harris,

202
Before starting the data analysis, three processing steps were required ( Figure 2). First, 203 some of the selected RSS news items were not worth analyzing because they were totally devoid 204 of information, simply advertising or summarizing a heterogeneous set of news of the day.

205
These items were deleted from the corpus. Second, the initial database continuously collects 206 RSS items on newspaper websites, and a similar item can be published several times without 207 changes. Therefore, we had to delete all the duplicate items (items with the same title and text).

208
During these two processing steps, more than 60 000 news items were deleted. After the 209 cleaning, the dataset contains 320 888 news items. To build the joint corpus (called EQ-MEDIA 210 in the following), we then enriched the news media dataset with a tagging process in two steps: 211 1) the geographical tagging of all mentioned countries using word dictionaries and 2) the 212 thematic tagging of all news mentioning a seismic event using an 'earthquake dictionary'. The  To more closely examine our dataset, we adopted a method inspired by Cox et al. (2008) 250 who analyzed the print-news media coverage of the recovery process following a forest fire.

251
The first step is to conduct a careful analysis of the content of the news itself to identify thematic 252 patterns but also possible "textual silences", defined by Huckin (2002) as "the omission of some 253 piece of information that is pertinent to the topic at hand". As we are dealing with thousands of 254 news items, this qualitative approach is complemented by a quantitative analysis based on

278
We classify the most frequently used words of the 'earthquake news' into one of these 279 categories of content and build two keyword dictionaries: a discourse content dictionary 280 corresponding to the above categories (table 1) and an identity matrix dedicated to actors (table   281 2). For this work to be manageable in a reasonable time, we adopt a threshold of a minimum of  Figure   307 3) share the remaining 28.6% of the coverage.

309
The curves of coverage intensity exhibit a similar trend for all earthquakes: the initial peak 310 is followed by an exponential decrease. This signature has been proved as typical of the media 311 coverage of dramatic events, characterized by an initial shock to public opinion (Boomgaarden,  percentages are partially due to the small numbers of keywords identified for each 356 of these themes, but it is the low frequency of these themes in the database that 357 prevented us from identifying more keywords.

358
It is interesting to note that the big aftershock of May 12 th in Nepal triggered a new cycle 359 of information. Although characterized by a peak of smaller intensity, the news content 360 followed a similar sequence to the one triggered by the main shock.

373
To give a sense of the framing of 'earthquake news', in the following, we build an (artificial 374 but well-informed) example of the evolution of the news content over time after an event. Of 375 course, there are to be variations due to elements of context, but our guess is that the main trends    Interestingly, that initial phase of coverage is also the phase with most scientific content.

406
The extensive use of the notion of magnitude, although often made at the expense of the notion 407 of seismic intensity, testifies to the successful transfer of a geophysical notion to the lay public.

408
We should also outline here that aftershocks are sometimes treated as singular events by the 409 press, with the notion of a seismic crisis remaining unclear to many. Among the most cited

496
We note a tendency of the international news to glorify the contribution of the international 497 community to help the 'poor and vulnerable'.  27.7% of the news mentions state representatives who are responsible for organizing the 571 public response, but regional and local public services are absent ( Figure 5). Surprisingly, only 572 8% of the news refers to civil and military security services and 7.7% to rescuers in general.

629
Although unprecedented, we are aware that our study also has some caveats. The use of 630 keywords to quantify themes and topics provides robust conclusions but is not completely 631 satisfactory. We tried to get around its limitations by preselecting words from a list of the most 632 frequently used terms. A further step is to engage with more complete techniques of text 633 analysis combining inductive and deductive approaches. We could, for example, use machine 634 learning methods such as word2vec (Le & Mikolov, 2014) for the simplification of the 635 collection of keywords and the quantification of the different steps of the news coverage.

636
However, this tool would complement but not replace the qualitative analysis of the content we 637 undertook in this study.

638
One of our working hypotheses was to demonstrate the existence of a global framing of 639 earthquake news and, to reach that goal, we chose to work on the international news, but it 640 would be important to undertake a similar analysis on the national and regional press as well as