Since 1996, death sentences in America have declined by more than 60 percent, reversing a generation-long trend toward greater acceptance of capital punishment. In theory, most Americans continue to support the death penalty. But it is no longer seen as a theoretical matter. Prosecutors, judges, and juries across the country have moved in large numbers to give much greater credence to the possibility of mistakes - mistakes that in this arena are potentially fatal. The discovery of innocence, documented in this book through painstaking analyses of media coverage and with newly developed methods, has led to historic shifts in public opinion and to a sharp decline in use of the death penalty by juries across the country. A social cascade, starting with legal clinics and innocence projects, has snowballed into a national phenomenon that may spell the end of the death penalty in America.

We describe the first version of the Media Frames Corpus: several thousand news articles on three policy issues, annotated in terms of media framing. We motivate framing as a phenomenon of study for computational linguistics and describe our annotation process.

Studies of political attention often focus on attention to a single issue, such as front-page coverage of the economy. However, examining attention to a single issue without accounting for the agenda as a whole can lead to faulty assumptions. One solution is to consider the diversity of attention; that is, how narrowly or widely attention is distributed across items (e.g., issues on an agenda or, at a lower level, frames in an issue debate). Attention diversity is an important variable in its own right, offering insight into how agendas vary in their accessibility to policy problems and perspectives. Yet despite the importance of attention diversity, we lack a standard for how best to measure it. This paper focuses on the four most commonly used measures: the inverse Herfindahl Index, Shannon's H, and their normalized versions. We discuss the purposes of these measures and compare them through simulations and using three real-world datasets. We conclude that both Shannon's H and its normalized form are better measures, minimizing the danger of spurious findings that could result from the less sensitive Herfindahl measures. The choice between the Shannon's H measures should be made based on whether variance in the total number of possible items (e.g., issues) is meaningful.

No abstract

Automated text analysis methods have made possible the classification of large corpora of text by measures such as topic and tone. Here, we provide a guide to help researchers navigate the consequential decisions they need to make before any measure can be produced from the text. We consider, both theoretically and empirically, the effects of such choices using as a running example efforts to measure the tone of New York Times coverage of the economy. We show that two reasonable approaches to corpus selection yield radically different corpora and we advocate for the use of keyword searches rather than predefined subject categories provided by news archives. We demonstrate the benefits of coding using article segments instead of sentences as units of analysis. We show that, given a fixed number of codings, it is better to increase the number of unique documents coded rather than the number of coders for each document. Finally, we find that supervised machine learning algorithms outperform dictionaries on a number of criteria. Overall, we intend this guide to serve as a reminder to analysts that thoughtfulness and human validation are key to text-as-data methods, particularly in an age when it is all too easy to computationally classify texts without attending to the methodological choices therein.