How the News Media Activate Public Expression and Influence National Agendas1 GaryKing2 InstituteforQuantitativeSocialScience HarvardUniversity UniversityofToronto,DepartmentofPoliticalScience,1/12/2018 1BasedonjointworkwithBenjaminSchneerandArielWhite(Science2017) 2GaryKing.org 1/23 Introduction ResearchDesign MainResults SupportingAnalyses Implications Introduction 2/23 • Randomization: usuallyimpossible • Endogeneity: mediaoutletscompeteforreaders • Spillover: 1interventionmayaffectallpotentialsubjects • NewTVtower. Somebehindhill,inradioshadow • Before/afterstudiesof“surprise”mediaevents • RolloutofFoxNewstosometownsandnotothers • Manyothers… • Assumptions: better,butunavoidablydubious ⇝“Profoundbiases,”>600%differencefromtruth • Estimands: different,ofsometimesquestionablerelevance • StatisticalProblems • CleverResearchDesigns(tryingtoapproximate randomization) • Butwestillcan’trandomize Statistical Problems: We Can’t Randomize Introduction 3/23 • NewTVtower. Somebehindhill,inradioshadow • Before/afterstudiesof“surprise”mediaevents • RolloutofFoxNewstosometownsandnotothers • Manyothers… • Assumptions: better,butunavoidablydubious ⇝“Profoundbiases,”>600%differencefromtruth • Estimands: different,ofsometimesquestionablerelevance • Randomization: usuallyimpossible • Endogeneity: mediaoutletscompeteforreaders • Spillover: 1interventionmayaffectallpotentialsubjects • CleverResearchDesigns(tryingtoapproximate randomization) • Butwestillcan’trandomize Statistical Problems: We Can’t Randomize • StatisticalProblems Introduction 3/23 • NewTVtower. Somebehindhill,inradioshadow • Before/afterstudiesof“surprise”mediaevents • RolloutofFoxNewstosometownsandnotothers • Manyothers… • Assumptions: better,butunavoidablydubious ⇝“Profoundbiases,”>600%differencefromtruth • Estimands: different,ofsometimesquestionablerelevance • Endogeneity: mediaoutletscompeteforreaders • Spillover: 1interventionmayaffectallpotentialsubjects • CleverResearchDesigns(tryingtoapproximate randomization) • Butwestillcan’trandomize Statistical Problems: We Can’t Randomize • StatisticalProblems • Randomization: usuallyimpossible Introduction 3/23 • NewTVtower. Somebehindhill,inradioshadow • Before/afterstudiesof“surprise”mediaevents • RolloutofFoxNewstosometownsandnotothers • Manyothers… • Assumptions: better,butunavoidablydubious ⇝“Profoundbiases,”>600%differencefromtruth • Estimands: different,ofsometimesquestionablerelevance • Spillover: 1interventionmayaffectallpotentialsubjects • CleverResearchDesigns(tryingtoapproximate randomization) • Butwestillcan’trandomize Statistical Problems: We Can’t Randomize • StatisticalProblems • Randomization: usuallyimpossible • Endogeneity: mediaoutletscompeteforreaders Introduction 3/23 • NewTVtower. Somebehindhill,inradioshadow • Before/afterstudiesof“surprise”mediaevents • RolloutofFoxNewstosometownsandnotothers • Manyothers… • Assumptions: better,butunavoidablydubious ⇝“Profoundbiases,”>600%differencefromtruth • Estimands: different,ofsometimesquestionablerelevance • CleverResearchDesigns(tryingtoapproximate randomization) • Butwestillcan’trandomize Statistical Problems: We Can’t Randomize • StatisticalProblems • Randomization: usuallyimpossible • Endogeneity: mediaoutletscompeteforreaders • Spillover: 1interventionmayaffectallpotentialsubjects Introduction 3/23 • Assumptions: better,butunavoidablydubious ⇝“Profoundbiases,”>600%differencefromtruth • Estimands: different,ofsometimesquestionablerelevance • NewTVtower. Somebehindhill,inradioshadow • Before/afterstudiesof“surprise”mediaevents • RolloutofFoxNewstosometownsandnotothers • Manyothers… • Butwestillcan’trandomize Statistical Problems: We Can’t Randomize • StatisticalProblems • Randomization: usuallyimpossible • Endogeneity: mediaoutletscompeteforreaders • Spillover: 1interventionmayaffectallpotentialsubjects • CleverResearchDesigns(tryingtoapproximate randomization) Introduction 3/23 • Assumptions: better,butunavoidablydubious ⇝“Profoundbiases,”>600%differencefromtruth • Estimands: different,ofsometimesquestionablerelevance • Before/afterstudiesof“surprise”mediaevents • RolloutofFoxNewstosometownsandnotothers • Manyothers… • Butwestillcan’trandomize Statistical Problems: We Can’t Randomize • StatisticalProblems • Randomization: usuallyimpossible • Endogeneity: mediaoutletscompeteforreaders • Spillover: 1interventionmayaffectallpotentialsubjects • CleverResearchDesigns(tryingtoapproximate randomization) • NewTVtower. Somebehindhill,inradioshadow Introduction 3/23 • Assumptions: better,butunavoidablydubious ⇝“Profoundbiases,”>600%differencefromtruth • Estimands: different,ofsometimesquestionablerelevance • RolloutofFoxNewstosometownsandnotothers • Manyothers… • Butwestillcan’trandomize Statistical Problems: We Can’t Randomize • StatisticalProblems • Randomization: usuallyimpossible • Endogeneity: mediaoutletscompeteforreaders • Spillover: 1interventionmayaffectallpotentialsubjects • CleverResearchDesigns(tryingtoapproximate randomization) • NewTVtower. Somebehindhill,inradioshadow • Before/afterstudiesof“surprise”mediaevents Introduction 3/23