Will 2012 Be A Low-Turnout Election? Some Evidence from Google Search Activity

by John Sides on October 25, 2012 · 8 comments

in Campaigns and elections,IT and politics

This is a guest post by political scientist Jesse Richman.

*****

This year one of my favorite informal election prediction metrics has been silent: the pizza indicator.  In nearly every election year since 2006 either the college Democrats or the college Republicans post signs by the elevators in the Arts and Letters building inviting students to come to a meeting and offering free pizza.  The Party offering pizza was the one that won the election in 2006, 2008, 2009 and 2010.  (2007 and 2011 were lower turnout state legislative election years with little pizza to be had).  This year neither organization is offering pizza, and there has been precious little paper spent even advertising their meetings.

Pizza aside, there are more powerful suggestions that 2012 will be a low turnout election.  I focus here on the frequency of Google searches for election-related information.  Searches for “vote” should be an indicator of interest in electoral participation.  Individuals might search for information on how to vote, or for information on voter registration, and other related information.  Therefore, searches for “vote” in Google should provide an indicator of the intensity of citizen interest in voting across states and across elections.  More frequent searches for “vote” should predict higher electoral participation.

I tabulated searches for “vote” from September through approximately Election Day in each state for the 2004, 2006, 2008 and 2010 U.S. elections, and compared search volume with state level voter turnout.  This comparison is of course only as good as Google’s algorithms that attempt to standardize search volume across time on the same 0 to 100 scale.

The graph below shows the relationship between voter turnout and searches for “vote” in the United States as measured by Google Insights for Search.  Overall, the relationship is strong.  States in which more people searched for “vote” in a particular election year had higher turnout.

The relationship is also evident after taking account of election year (midterm elections are clustered in the lower left hand portion of the chart).  Once election year is taken into account, a shift from 10 to 20 on the search scale is associated with a four percent increase in voter turnout.

If searches for “vote” do predict electoral participation, then 2012 should have lower turnout than 2008 or 2004.  The next figure shows nationwide searches for “vote” in 2004, 2008 and (so far) in 2012.  (Note that week one is the first week that includes any dates from a given year so week one in 2004 started two days before week one in 2008 and four days earlier than week 1 in 2012, which should bias the 2012 results toward higher numbers as of this writing.)

2012 had lower search frequencies in September at October compared with the last two election years.   This implies that if the present trends continue 2012 will also have lower voter turnout.

I also collected the search frequency for “vote” in several swing states (Colorado, Florida, Iowa, Michigan, Nevada, North Carolina, Ohio, Pennsylvania, Virginia, Wisconsin) from September through mid-October.  In no instance was the frequency of searches for “vote” higher in 2012 than for previous years.  In every case it was substantially lower.  Often the search volume was only half of the levels experienced in early election seasons.

Searches for the phrase “register to vote” show the same pattern:

Interest in voter registration peaks in early October, around the time that many states have voter registration deadlines.  In 2012 that peak was much lower (58) relative to 2008 (90) and 2004 (100).

The relatively higher internet use among younger voters (and their pronounced tendency to favor Obama) mean that declining search frequency should give particular pause for Democrats.  It suggests that interest in this election is lower than in the last two presidential cycles and that turnout will be lower as well.  Perhaps some pizza is called for.

{ 8 comments }

fede October 25, 2012 at 3:53 pm

Does the relationship in Figure 1 hold even when you center around the election mean–or at least the grand means of midterm and presidential elections?

Jesse October 26, 2012 at 12:41 pm

I ran a regression with fixed effects by year. I think this has the effect of centering around the election mean. And it does hold in the sense of remaining statistically significant.

thegalen October 25, 2012 at 4:36 pm

According to this, 42 million newly registered were added to the rolls in 2008 as compared to 2004. Only 13 million for 2004 as compared to 2000. Couldn’t the drop off simply be a mechanical effect of so many people already being registered? That is, if you’re already registered, why would you Google “register to vote”?
http://www.wired.com/threatlevel/2008/11/nationwide-vote/

B.G. Klin October 25, 2012 at 4:38 pm

The lurking question is whether internet search and use patterns are different today than in 2004 and 2008. The rise of social networks, Twitter and even the increased use of mobile internet devices would seem to suggest that today Google searches may not be quite the indicator they were four or eight years ago. It may be that today people are getting their voting information from online sources that are not reflected in the data analyzed here. For instance, it may be that political chatter and information exchange on Facebook has displaced some of the old Google search traffic. If that is true, then election interest could be higher than the Google data would appear to suggest.

Bob"Romney writes Cocession already" Jones October 25, 2012 at 4:56 pm

So you’re saying Turnout and searches were highest in 2004? which by every available metric was a low turnout election.

Seems like you need to re-write this article using a better model, it is poorly written and not thought out correctly.

Why not use keywords like Obama, Romney, polls etc.

And the poster above is correct, no one uses Google anymore.

Nadia Hassan October 26, 2012 at 8:56 am

2004 was a high turnout election.

Jesse October 26, 2012 at 12:52 pm

I’m not sure it’s accurate to say that “no one uses Google any more”. Its search market share is at 66 percent, just off its all time high. http://www.webpronews.com/bing-up-google-down-in-latest-search-market-report-2012-09.
That said, the increased use of mobile devices and social networks do potentially complicate the validity of this measure across time. It’s something I’d like to try to get a better handle on. I think mobile devices using Google search would still show up in this search data.

Jesse January 9, 2013 at 3:47 pm

Post-election… it looks like the search volume was (on the whole) an indicator of lower turnout. According to http://elections.gmu.edu/Turnout_2012G.html turnout VEP turnout for the United States as a whole was down. On average across the states I examined in this piece (DE, ND, and WY were excluded because they lacked weekly search volumes) turnout was down by 3.3 percent and search frequency was down by 2.02 points on Google’s 0 to 100 frequency scale.

Comments on this entry are closed.

Previous post:

Next post: