By some accounts, the American public is awash in misinformation. Polarization, selective exposure, biased processing, and politicians and media outlets that seem ever readier to promote convenient fictions and deny inconvenient facts make it plausible. Yet there is also reason, in the underlying survey items, to suspect the headlines of being considerably overdrawn. Among other things, the questions on which they rest are frequently phrased as matters of opinion ("As far as you know ...," "Would you say that...”), and frequently lack explicit DK options and more generally invite guessing. We assemble a novel corpus of media polls to estimate the frequency of such “inflationary” features in misinformation items. We then use a series of survey experiments to estimate the effects of such question design features on the proportion of incorrect responses. We show that common features of misinformation items considerably overstate misinformation levels.
Robert C. Luskin, Gaurav Sood, Yul Min Park, and Joshua Blank