Leveraging Predictive Power to Estimate Intermedia Agenda Setting

Abstract

Over the last 50 years, studies in agenda setting, how news coverage influences perceptions of importance, and agenda building, how entities influence media coverage, have found evidence that different news media influence one another. However, there has been a dearth of studies evaluating and comparing the predictive power of different theoretical frameworks of news coverage. The thesis addresses the gap in comparing theories by leveraging estimated intermedia agenda setting effects from simplified theoretical assumptions to forecast news coverage and measure the error of those forecasts. The thesis begins by computing and evaluating univariate, bivariate, and multivariate SARIMAX models for coverage of 432 individuals from 12 top U.S. news websites, the U.S. president’s Twitter feed, and Google search trends, for the period from 2017-2018. The thesis then compares where different models outperform other models in the predictive accuracy of each model from each source. In line with previous studies in agenda setting using within-sample evaluation, this study finds evidence for reciprocal agenda setting effects between all three communications architectures when evaluating models on out-of-sample forecasts. Since at least some news coverage can be modeled and predicted from simple assumptions of agenda setting, it may be possible for journalism scholars to evaluate advancements in theory by measuring improvements in forecast accuracy.