Public opinion research comes to a narrow focus especially in presidential election years, and this year polling analytics got its fair share of attention. Mike Malecki, of the YouGov analytics team (and formerly a postdoc under Andy Gelman), will present some of the R code underlying online survey research. Pollsters gonna poll and aggregators gonna aggregate — in the end, YouGov hit the national vote split with almost no error using a large national sample (YG release; marginoferror; politico). We also partnered with Microsoft Research to do a daily survey on a massive self-selected panel, as well as live interactive polling during the debates on the Xbox Live platform (TIME blog; NPR; Huffington Post).
Mike will present three parts of these projects as a code review: (1) multilevel regression and poststratification to model party identification in the population; (2) an all-R client to consume data from a live-streaming web service and transform it from clickstream to crosstabs; and (3) a hidden-Markov-inspired approach to individual-level change over time on repeated questions.
Pizza from a new place starts at 6:15, Mike will begin around 7 then we will head to a nearby bar.
A big "thank you" goes to Knewton for hosting us again.