Sequential Experiments and News Media Effects

Presentation Date: 

Tuesday, October 30, 2018


Harvard Psychology Graduate Student Methods Dinner

Presentation Slides: 

We report on the results of first large scale randomized news media experiment, with a special focus on the sequential experimental design we used. Instead of fixing your n ahead of time and finding out about your p-value post hoc, you can choose your p-value (or CI) ahead of time and discover the n needed. Your experiment should not risk collecting more data than necessary (wasting your time, grant resources, and research subjects' patence) or less than you need to draw firm conclusions.

Substantively, we demonstrate that even small news media outlets can cause large numbers of Americans to take public stands on specific issues, join national policy conversations, and express themselves publicly—all key components of democratic politics—more often than they would otherwise. After recruiting 48 mostly small media outlets, and working with them over 5 years, we chose groups of these outlets to write and publish articles on subjects we approved, on dates we randomly assigned. We estimate the causal effect on proximal measures, such as website pageviews and Twitter discussion of the articles’ specific subjects, and distal ones, such as national conversation in broad policy areas. Our intervention increased discussion in each broad policy area by a substantial ≈62.7% (relative to a day’s volume), accounting for 13,166 additional posts over the treatment week, with similar large effects across population subgroups. We also discuss the normative implications of this for individual journalists, the national ecosystem of media outlets, and democratic politics. This talk is based on work recently published in Science with Benjamin Schneer and Ariel White; for more information, see