Abstract
A media storm is a sudden surge in news coverage of an item, producing high attention for a sustained period. Our study represents the first multi-issue, quantitative analysis of storm behavior. We build a theory of the mechanisms that drive media storms and why the “anatomy” of media storms differs from that of non-storm coverage. Specifically, media storm coverage should change less explosively over time, but be more sharply skewed across issues, compared to non-storm coverage. We offer a new method of operationalizing media storms and apply our operationalization to U.S. and Belgian news. Even in these two very different cases, we find a common empirical storm anatomy with properties that differ from those of non-storm coverage in the predicted fashion. We illustrate the effects of media storms on the public through discussion of four key examples, showing that online search behavior responds strongly to media storms.
Notes
1. Subtopic coding yielded the following inter-coder reliability statistics for NYT: percentage agreement = 90.7%, Cohen’s kappa = 0.897, Krippendorff’s alpha = 0.898. For DS, percentage agreement = 65.0%, Cohen’s kappa = 0.643, Krippendorff’s alpha = 0.644.
2. Storms typically last only a couple of weeks (16 days on average for NYT) but rarely for a perfect multiple of 7 days. Thus, collapsing storm proportions on fixed calendar weeks in order to provide a more direct comparison with general coverage would inaccurately reflect the average volume of storm coverage. By contrast, we can calculate descriptive statistics for non-storm coverage based on fixed calendar weeks because dropping all storm stories does not interrupt the chronology of days for each data set, thereby allowing us to collapse non-storm stories by week. Because we are examining the average subtopic proportions during storms versus during non-storm weeks, the comparison is appropriate.
3. Note that the minimum proportion of attention per storm for both NYT and DS is listed below our 20% cut point because these descriptive statistics are calculated by averaging the total proportion of coverage for each storm across its entire period and then taking the mean of all these means. Because each storm is considered a storm as long as the current rolling 7-day period meets our 20% criterion, the overall mean proportion of coverage a subtopic receives across the full storm can be less than 20% (since strong days of coverage can fuel multiple 7-day rolling windows past the cut point, even if the total mean for all days is below it).
Additional information
Notes on contributors
Amber E. Boydstun
Amber E. Boydstun is an Assistant Professor in Political Science, University of California, Davis.
Anne Hardy
Anne Hardy is a PhD student in Political Science, University of Antwerp, Belgium.
Stefaan Walgrave
Stefaan Walgrave is a Professor in Political Science, University of Antwerp, Belgium.