Google searches and the U.S. presidential election: Part II
Why we need to rethink traditional polling
I published an analysis back in October in which I used Google searches to predict the results of the 2020 U.S. presidential election. I was largely focused on six swing states — Arizona, Florida, Michigan, North Carolina, Pennsylvania, and Wisconsin. This analysis was based upon the work of Seth Stephens-Davidowitz, a data scientist with extensive experience working with Google search data.
Anyone who followed the election knows that the polls missed the mark in several key states, but how did the predictions gleaned from Google search data hold up?
Part 1: Candidate Name Order
The first part of my analysis was based upon Stephens-Davidowitz’s finding that when individuals make Google searches containing the names of two candidates, such as “Trump Clinton debate” or “Clinton Trump debate,” they tend to subconsciously place the name of the candidate they will vote for first.
In 2016, however, Stephens-Davidowitz identified Trump as a unique candidate. He found that the rules he had previously identified for candidate name order in Google searches had broken down, and that almost every state had “Trump” ordered first in more searches that contained both…
