Belief Network Timeseries Analaysis

Last year I came across a working paper for AJS on Belief Network Analysis by Andrei Boutyline [1]. The paper looks at American National Election Survey data and examines two theories for the process of political opinion formation: Lakoff’s Theory of Moral Politics and Campbell’s Theory of Political Identity. This project, in collaboration with Sujaya Maiyya, was focused on extending BNA to the American National Election Survey timeseries data to test some of the claims Andrei made in response to an investigation from Delia Baldassarri using Relational Class Analysis [2]. The original work was performed by analyzing survey data from the year 2000, but we argue that no claims can be made about this process unless we make a longitudinal investigation.

The figure above is our visualized result from the year 2000, which closely matches Andrei’s results in [1]. The visualization essentially shows Ideological Identity to be towards the center of the network and some items like Gender equality and Crime spending towards the far periphery, indicating that they are loosely correlated to the other survey questions. We think the visualization is fascinating because a quick glance at the centrality in this picture tends to closely match results from the path analysis used to test the theories. Shortest-path betweenness was used as the ‘importance’ measure in these systems, and Ideological identity indeed came out as the ‘most important’.

One of the most interesting questions posed in the original BNA paper was the heterogeneity in US sub-populations by race, gender, income, and other factors. Aside from robustness measurements, Andre found that high-income, the male gender, and ethnically white sub-populations tended to have a higher average correlation (measured on separate axes). This could be interpreted as a measure of consistency across survey questions that correspond to more firm political left-right positions. In our follow-on, we decided to plot the gender and income axes as they changed over time. The results are in the following two figures:

The simplest observation is that the sharp increase in mean constraint starting in the 2008 election is evidence of a more contentious political sphere in the US – this matches findings from a broad range of other studies. We expect that the global variation in both the income and gender cases are likely due to fluctuation in data availability in those survey years. Not all questions were asked in all years, and thus data imputations will be needed for future work.

The fascinating part is that both high income and male respondents were consistently higher than low income and female respondents and the gap did not change since 1984. Why is this? Sociologists might say there is more social pressure towards male and high-income respondents to carry consistent political beliefs. Are inconsistent responses better perceived by members of both the affiliate party and the opposing party? Do less consistent beliefs relate to weaker argument formation in communal discussion spheres? These very preliminary results show promise but leave us simply with more questions than answers. After all further work is completed on this project, still more work will need to be done into how these results might manifest themselves in news media and individual rhetoric.

Although we were able to perform some simple analyses quickly, we face several future challenges in working with the network timeseries data. The following stepping stones need to be made before we can continue:

  • reasonable methods for data imputation for missing questions and missing survey years
  • regression or fit models that accurately describes how these belief networks change over time
  • explanations for how consistency shapes discourse
  • explanations for why gender and income are related to average constraint

If anyone has explanations for these findings please let me know: dcornell [at]

This project was completed as part of the UCSB IGERT Network Science Traineeship. We are required to do several network-related projects over the course of our two year traineeship.

[1] Boutyline, A., & Stephen Vaisey. (2015). Belief Network Analysis: A Relational Approach to Understanding the Structure of Attitudes. American Journal of Sociology, 53, 160.

[2] Baldassarri, D., & Goldberg, A. (2014). Neither Ideologues nor Agnostics: Alternative Voters’ Belief System in an Age of Partisan Politics. American Journal of Sociology, 120(1), 45–95.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s