New Paper Uses Google Web Search Data to Suggest EDR Could Have Added 3-4 Million Voters in 2012

[Image courtesy of oxfordjournals]

My friend and colleague Mike Alvarez of CalTech shared a new paper appearing in Political Analysis yesterday that not only has interesting conclusions about the effect of registration deadlines but also suggests that readily-available but under-appreciated data on web searches could help us get a better handle on how voters perceive the election process.

The paper, “Estimating Voter Registration Deadline Effects with Web Search Data” by Alex Street, Thomas Murray, John Blitzer and Rajan Patel, dives into the search data available via the Google Trends website and examines when in 2012 people searched for voter registration in comparison to registration deadlines. Here’s the abstract:

Electoral rules have the potential to affect the size and composition of the voting public. Yet scholars disagree over whether requiring voters to register well in advance of Election Day reduces turnout. We present a new approach, using web searches for “voter registration” to measure interest in registering, both before and after registration deadlines for the 2012 U.S. presidential election. Many Americans sought information on “voter registration” even after the deadline in their state had passed. Combining web search data with evidence on the timing of registration for 80 million Americans, we model the relationship between search and registration. Extrapolating this relationship to the post-deadline period, we estimate that an additional 3-4 million Americans would have registered in time to vote, if deadlines had been extended to Election Day. We test our approach by predicting out of sample and with historical data. Web search data provide new opportunities to measure and study information-seeking behavior.

The authors do make some important assumptions – most notably, that searching for “voter registration” after the deadline is just as strong evidence of desire to register as in the pre-deadline period. They conclude, however, that the assumption is plausible and use it to get a sense of how many voters may have missed the deadline in 2012 despite their search-indicated interest in registering, Using those numbers, they conclude:

Summing across states that did not allow EDR in 2012, our models suggest that around 3.5 million people would have registered in the post-deadline period, if this had been possible… This would have added 2% points to the total number registered nationwide. High turnout among late registrants, and full turnout among those who register on Election Day, implies that 80% or more of these people would have voted, producing a 3% point increase in turnout… [internal citations omitted].

One point that isn’t clear (made by Florida’s Michael McDonald yesterday on Twitter) is how many of these searches are evidence of “portable registration”; i.e., people who are already registered in one place but are moving to another. Assuming these searches led to new registrations, they would not be an indication of a net increase in the total number of voters. Still, the use of web searches to size the problem is not only interesting but important.

The paper goes on to discuss how this same methodology – “search forensics”, if you will – could be used in other contexts:

An obvious extension of our work would be to use web search data to study other aspects of election administration. The effects of electoral rules have acquired new relevance in the wake of the Supreme Court decision in Shelby County v. Holder (2013), which invalidated a key section of the 1965 Voting Rights Act, allowing many more jurisdictions to amend electoral rules without federal oversight. As Kousser and Mullin (2007) observe, the United States is cross-cut by boundaries for elections at the local, state, and federal levels. Besides variation in registration deadlines, U.S. elections differ in myriad ways, such as the presence or absence of voter ID laws (Bentele and O’Brien 2013), the availability of mail ballots, sample ballots, and other information (Wolfinger, Highton, and Mullin 2005; Kousser and Mullin 2007), or the accessibility of polling places (Brady and McNulty 2011). The onus is on the voter to find out about the details, and searching online is now the obvious way to do so. [emphasis added]

This is a fascinating and important area of inquiry going forward; as more and more individuals use search engines to locate information, data on those searches can help election officials understand what their voters want – and want to know.

Thanks to the authors for their work – and to Mike Alvarez for sharing the paper!