Electoral Chairs’ Seminars – 4 May

Mixed modes of data collection and administration and new sources of samples in the US 2020 election. How did they fare?

Claire Durand (Université de Montréal)
Tim. P. Johnson (University of Chicago)
Luis P. Pena Ibarra (Université de Montréal)

This paper presents the results of an analysis of the performance of the polls of the campaign in the 2020 U.S. Presidential election. We first focus on mixed mode and estimate the trends in support for Joe Biden according to the use of mixed mode. We validate these trends statistically using a longitudinal multilevel analysis where polls are nested within pollsters. We then focus on the last ten days of the campaign and observe the overwhelming presence of mixed mode polls among the best performing polls. This analysis allows for observing also that the best performing web polls usually do not use Web Opt-in panel. This leads us to validate an alternate multilevel model, which shows that the combination of the use of mixed mode and the use of other sampling sources than Web panels leads to better estimates of support for Biden. We conclude that we will need to examine more seriously the methods used by pollsters, particularly web pollsters, and differentiate the polls according to criteria that are more relevant than the unique mode of administration or data collection. 

Contact Semih Çakır if you would like to participate in the seminar.

This content has been updated on 3 May 2022 at 10 h 12 min.