Can Google Search Behavior Predict Political Behavior?

One of the driving forces behind the creation of Google Insights was the observation that Google searches can predict flu epidemics more quickly than other types of observations. Shauna Reilly, Sean Richey, and Benjamin Taylor have a forthcoming article (nongated), which suggests that political behavior too can be predicted from search terms. In particular they find that the level of Google searches on ballot questions taken one week before Election day correlates with actual participation on those ballot measures.

The authors highlight that this illustrates that measures based on Google search data may be valid measures of behavioral intentions (see yesterday’s post on racially charged search terms). Yet, it also suggests that search data can be used for forecasting. just for fun, I created a simple chart (sorry, can’t figure out how to embed it) for search behavior for Gingrich. It seems to track changes in the polls pretty well: Gingrich first started making inroads in polls taken November 4 and 5. Of course, polling data is only available several days later. I am sure more sophisticated exercises could be pursued.

Update: Lori Williams pointed me to this excellent visualization of Google search trends and poll numbers for Republican candidates (the Perry and Cain ones are especially suggestive).

5 Responses to Can Google Search Behavior Predict Political Behavior?

  1. Josh Putnam December 1, 2011 at 12:19 pm #

    There’s a strange lag of a week at the moment, but Google Trends output is embeddable. See here: http://frontloading.blogspot.com/2009/05/2012-gop-candidate-emergence-tracker.html

    Feel free to grab the HTML code from that post and edit the names as needed.

  2. Max Hell December 1, 2011 at 1:28 pm #

    On a related note, Oxford grad student Jon Mellon has a paper on Google Trends. Here’s the abstact and a link:

    “Search Indices and Issue Salience: the properties of Google Trends as a measure of issue salience. Measures of issue salience are widely used in political science, particularly within agenda setting. However progress is restricted due to limited data availability as the commonly used “most important problem” question is only published monthly for a handful of countries. Internet searches could act as measures of issue salience in a population if they reflect public motivation to search for information on a topic. Google Trends provides weekly indices of internet searches for every country and topic. This paper contributes a detailed approach to assessing internet search index validity. Validity is assessed qualitatively by examining search terms within a search index and quantitatively by comparing search indices to Gallup’s “most important problem” measure of issue salience. The salience of four issues, macroeconomics, terrorism, immigration and fuel prices, can be measured in the US using search indices. Weekly measures of issue salience are generated for these issues from 2004-2010.”
    http://www.sociology.ox.ac.uk/documents/working-papers/2011/swp1101.pdf

  3. Joe Ripberger December 1, 2011 at 1:43 pm #

    Thanks for the excellent post…google search is a tremendous tool that social scientists are only beginning to tap. For more on google search trends as a measure of public attentiveness, check out “Capturing Curiosity: Using Internet Search Trends to Measure Public Attentiveness” which was recently published in Policy Studies Journal. (SSRN: http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1539137).

    On google search trends and political behavior, Colin Swearingen and I have a working paper called “Examining the Effect of Public Attention on U.S. Senate Election Outcomes” (SSRN: http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1924574). Using data from Senate elections between 2004-2010, we find that relative attention (as measured by google searches) is a useful predictor of electoral outcomes–especially in open seat elections.

  4. Eric Hines December 1, 2011 at 3:06 pm #

    A fascinating comparison is to look at searches for multiple GOP candidates and President Obama. Herman Cain’s rise and fall is very clear in the data, but Gingrich’s rise as the anti-Romney is not as sudden nor really significant outside of a handful of states.

  5. Mike Danielson December 2, 2011 at 1:58 pm #

    The spike in Rick Santorum searches last May, of course, has to do with the fact that John Stewart suggested folks google Santorum. It’s noteworthy that his poll numbers reached their peak not too long after that, even though the connection is indirect at best. Maybe his poll bump had to do with some sympathy support–even though these are probably not the same people who were googling “santorum”.