2020 U.S. Election Digital Trace Project

• In Event: Forecasting Elections

Abstract

The 2020 Digital Trace project is an extension of a study that we conducted during the 2018 U.S. mid-term elections – results are published in a forthcoming article in IJF. We test a regression and post-stratification strategy that combines samples of digital traces with a stratification frame containing individual-level socio-economic data, in order to generate area forecasts of the two-party vote of Democrats and Republicans in the 2018 Texas congressional district and Senate election. Our implementation assumes we can observe, and sample, individuals signaling their preference by favoring one virtual location over another. In our case, visiting Democrat versus Republican Facebook pages provided accurate digital measures of voter preferences.

For the 2020 U.S. Presidential Elections we implemented a modified national version of the 2018 Texas study we conducted. The goal was similar: generating accurate forecasts, at the congressional district level, of votes for Democrat and Republican congressional candidates and votes for the Presidential contenders. There are three key data components to the project: 1) collection of national digital trace data from diverse social media – employing a similar methodology described in the IJF paper; 2) conducting individual-level surveys using online subject pools – specifically the CESS Online U.S. subject pool and commercial crowd-sourced workers (we recently implemented a similar data collection strategy for the recent Indian elections); 3) commercial voter file data bases such L2. Complementing this multi-mode data collection strategy that developed innovative machine learning techniques for conducting MRP estimations of vote preferences at the congressional district area level. These build on models employing the individual data we collected and a stratification frame developed from the commercial voter files and census data.

We generated weekly forecasts of 2020 U.S. congressional seat votes and Presidential votes for the Democrat and Republican contenders.

Author

• Raymond Duch, CESS Nuffield