Googling Politicians: A Measure of Domain Concentration in Search Results

• In Event: Poster Session: New Research in Information Technology and Politics & Canadian Politics
  In Poster Session: Poster Session: New Research in Information Technology & Politics

Abstract

Democracy cannot be achieved without citizens being well-informed of authentic information through multiple sources. In the digital era, most people consume political information via online channels, the largest of which being Google. This has generated concern over the potential for “filter bubbles” resulting from personalized search results and concentrated web sources. Although many scholars have studied the influence of personalization, relatively fewer studies have explored to which extent Google search results are concentrated, meaning that a few dominant websites are more likely to be present in the search results and ranked on the top. The question, whether this concentration is influenced by political biases, has not been answered. This work fills in the gap by measuring domain concentration of Google search result pages (SERPs) and its correlation with partisanship.
We have been regularly collecting Google SERPs for 604 queries – including the names of presidential candidates, key figures in presidential administrations, and members of Congress – since January 1st, 2020 using a custom Python package, WebSearcher. Searches are initiated by proxy from each of 435 congressional districts (plus the District of Columbia) in order to capture potential differences in search results by location.
We intend to answer the following questions through statistical analysis. First, we want to understand how much concentration is there in Google search results. Specifically, we investigate whether Google provides multiple sources instead of a few well-known sources. We then evaluate the effect of the “filter bubble”, where searches provide information from liberal sources regarding liberal politicians or conservative sources for conservative politicians, as well as the extent to which search results for representatives differ within versus outside of the areas they represent. Finally, we test whether Google shows content from any misinformation sources, and if so, whether there is any pattern regarding which searches for which politicians evokes misinformation sources. Preliminary findings indicate that top search results tend to concentrate on a few dominant domains (e.g. wikipedia.org), consistent with previous works. This provides evidence to the importance of studying google search algorithms and frames a measurement of biases of political information channels.

Authors

• Zhen Guo, Northeastern University

• David Lazer, Northeastern University

• Jon Green, Duke University

• Ronald Robertson, Northeastern University

• Allison Wan

• Alyssa Hasegawa Smith, Northeastern University