Fine-Grained Analysis of Diversity Levels in the News

ABSTRACT

Many researchers consider the presentation of diverse content as a prerequisite for the news media to fully exercise their democratic mandate. While prior news diversity studies have contributed important theoretical insights, we argue here that scholarly knowledge of this concept can be significantly advanced by employing computational methods for text analysis. Using automated methods, researchers can increase both the scope of data being analyzed and the resolution of the analysis. This article presents a novel framework for analyzing news diversity consisting of two distinct stages. In the first stage, a computational text classification method is used to analyze, at a high resolution, the attention given in news texts to a broad range of political and social issues. In the second stage, the text classifications are aggregated, and the distributions of media attention to those issues (i.e., news diversity) are assessed on a large scale. After presenting the novel approach, we illustrate its usefulness for testing theoretical hypotheses about news diversity. We compare the diversity of economic coverage in three elite and three popular US newspapers (N = 252,807 articles) and find that a fine-grained analysis relaxes concerns raised in previous studies about low content diversity in the popular press.

Disclosure Statement

No potential conflict of interest was reported by the authors.

Notes

1. We give all sentences the same weight in the analysis regardless of their length. While one might argue that longer sentences should be given more weight because they might discuss topics in greater detail, the literature suggests that longer sentences are not more effective than shorter ones in communicating information. In fact, for most people, longer sentences are more difficult to comprehend (Mikk, 2008; Thorton et al., 2000).

2. We focus on economic issues, broadly speaking. While many of the topics we analyze are politically relevant (as shown in the empirical illustrations below), not all are directly related to the political domain.

3. This list does not contain all aspects of the economy that can possibly appear in the news. Rather, it represents economic sub-domains featured prominently in English-speaking newspapers in the two decades under investigation (1995–2017).

4. We consider newspaper articles with entropy scores of 0.35 and 0.42 as high on diversity (see Table 3) only after having inspectedentropy scores in a large sample of articles. Scores of 0.35 and 0.42 are well above the average diversity in both a New York Times article (M = 0.2, SD = 0.18) and a USA Today article (M = 0.19, SD = 0.17).

5. Although the differences between elite and popular newspapers are statistically significant at the week, month, and quarter levels, they are substantively very small in all cases (see Table 4).