Skip to main content

Inferring Population Preferences via Mixtures of Spatial Voting Models

  • Conference paper
  • First Online:
Social Informatics (SocInfo 2016)

Part of the book series: Lecture Notes in Computer Science ((LNISA,volume 10046))

Included in the following conference series:

Abstract

Understanding political phenomena requires measuring the political preferences of society. We introduce a model based on mixtures of spatial voting models that infers the underlying distribution of political preferences of voters with only voting records of the population and political positions of candidates in an election. Beyond offering a cost-effective alternative to surveys, this method projects the political preferences of voters and candidates into a shared latent preference space. This projection allows us to directly compare the preferences of the two groups, which is desirable for political science but difficult with traditional survey methods. After validating the aggregated-level inferences of this model against results of related work and on simple prediction tasks, we apply the model to better understand the phenomenon of political polarization in the Texas, New York, and Ohio electorates. Taken at face value, inferences drawn from our model indicate that the electorates in these states may be less bimodal than the distribution of candidates, but that the electorates are comparatively more extreme in their variance. We conclude with a discussion of limitations of our method and potential future directions for research.

This is a preview of subscription content, log in via an institution to check access.

Access this chapter

Chapter
USD 29.95
Price excludes VAT (USA)
  • Available as PDF
  • Read on any device
  • Instant download
  • Own it forever
eBook
USD 39.99
Price excludes VAT (USA)
  • Available as EPUB and PDF
  • Read on any device
  • Instant download
  • Own it forever
Softcover Book
USD 54.99
Price excludes VAT (USA)
  • Compact, lightweight edition
  • Dispatched in 3 to 5 business days
  • Free shipping worldwide - see info

Tax calculation will be finalised at checkout

Purchases are for personal use only

Institutional subscriptions

Similar content being viewed by others

Notes

  1. 1.

    Code and data for analyses are available at https://github.com/anahm/inferring-population-preferences.

References

  1. Abramowitz, A.I., Saunders, K.L.: Is polarization a myth? J. Polit. 70(2), 542–555 (2008)

    Article  Google Scholar 

  2. Airoldi, E.M., Blei, D.M., Fienberg, S.E., Xing, E.P.: Mixed membership stochastic blockmodels. In: Advances in Neural Information Processing Systems, pp. 33–40 (2009)

    Google Scholar 

  3. Ansolabehere, S., Palmer, M., Lee, A.: Precinct-level election data (2014). http://hdl.handle.net/1902.1/21919

  4. Ansolabehere, S., Pettigrew, S.: Cumulative CCES Common Content (2006–2012) (2014). doi: 10.7910/DVN/26451

    Google Scholar 

  5. Barberá, P.: Birds of the same feather Tweet together: Bayesian ideal point estimation using Twitter data. Polit. Anal. 23(1), 76–91 (2015)

    Article  Google Scholar 

  6. Bartels, L.M.: Beyond the running tally: partisan bias in political perceptions. Polit. Behav. 24(2), 117–150 (2002)

    Article  Google Scholar 

  7. Bonica, A.: Database on ideology, money in politics, and elections: Public version 1.0 (2013). http://data.stanford.edu/dime

  8. Bonica, A.: Mapping the ideological marketplace. Am. J. Polit. Sci. 58(2), 367–386 (2014)

    Article  Google Scholar 

  9. DiMaggio, P., Evans, J., Bryson, B.: Have american’s social attitudes become more polarized? Am. J. Sociol. 102, 690–755 (1996)

    Google Scholar 

  10. Ding, W., Ishwar, P., Saligrama, V.: Learning mixed membership mallows models from pairwise comparisons arXiv:1504.00757 (2015)

  11. Downs, A.: An economic theory of political action in a democracy. J. Polit. Econ. 65(2), 135–150 (1957)

    Article  Google Scholar 

  12. Enelow, J.M., Hinich, M.J.: The Spatial Theory of Voting: an Introduction. Cambridge University Press, Cambridge (1984)

    MATH  Google Scholar 

  13. Fiorina, M.P., Abrams, S.J.: Political polarization in the American public. Ann. Rev. Polit. Sci. 11, 563–588 (2008)

    Article  Google Scholar 

  14. Fiorina, M.P., Abrams, S.J., Pope, J.C.: Polarization in the American public: misconceptions and misreadings. J. Polit. 70(2), 556–560 (2008)

    Article  Google Scholar 

  15. Flaxman, S.R., Wang, Y.X., Smola, A.J.: Who supported Obama in 2012? Ecological inference through distribution regression. In: Proceedings of the 21st ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp. 289–298. ACM (2015)

    Google Scholar 

  16. Gerber, E.R., Lewis, J.B.: Beyond the median: voter preferences, district heterogeneity, and political representation. J. Polit. Econ. 112(6), 1364–1383 (2004)

    Article  Google Scholar 

  17. Gerrish, S., Blei, D.M.: Predicting legislative roll calls from text. In: Proceedings of the 28th International Conference on Machine Learning, pp. 489–496 (2011)

    Google Scholar 

  18. Krafft, P., Moore, J., Desmarais, B., Wallach, H.M.: Topic-partitioned multinetwork embeddings. In: Advances in Neural Information Processing Systems, pp. 2807–2815 (2012)

    Google Scholar 

  19. Lee, J.M.: Assessing mass opinion polarization in the U.S. using relative distribution method. Soc. Indic. Res. 124(2), 571–598 (2014)

    Article  Google Scholar 

  20. Levendusky, M.S., Pope, J.C., Jackman, S.D.: Measuring district-level partisanship with implications for the analysis of U.S. elections. J. Polit. 70(3), 736–753 (2008)

    Article  Google Scholar 

  21. Levendusky, M.S., Pope, J.C.: Red states vs. blue states going beyond the mean. Publ. Opin. Q. 75(2), 227–248 (2011)

    Article  Google Scholar 

  22. Lewis, J.B.: Estimating voter preference distributions from individual-level voting data. Polit. Anal. 9(3), 275–297 (2001)

    Article  Google Scholar 

  23. Lewis, J.B., DeVine, B., Pitcher, L., Martis, K.C.: Digital boundary definitions of United States congressional districts, pp. 1789–2012 (2013). http://cdmaps.polisci.ucla.edu

  24. McCarty, N., Poole, K.T., Rosenthal, H.: Polarized America: The Dance of Ideology and Unequal Riches, vol. 5. MIT Press, Cambridge (2006)

    Google Scholar 

  25. Poole, K.T., Rosenthal, H.: A spatial model for legislative roll call analysis. Am. J. Polit. Sci. 29(2), 357–384 (1985)

    Article  Google Scholar 

  26. Ruths, D., Pfeffer, J.: Social media for large studies of behavior. Science 346(6213), 1063–1064 (2014)

    Article  Google Scholar 

  27. Tausanovitch, C., Warshaw, C.: Measuring constituent policy preferences in congress, state legislatures, and cities. J. Polit. 75(2), 330–342 (2013)

    Article  Google Scholar 

  28. United States Census Bureau: Tigerweb state-based data files: Voting districts - Census 2010 (2010). http://tigerweb.geo.census.gov

  29. Zhang, Y., Friend, A., Traud, A.L., Porter, M.A., Fowler, J.H., Mucha, P.J.: Community structure in congressional cosponsorship networks. Phys. A 387(7), 1705–1712 (2008)

    Article  Google Scholar 

Download references

Acknowledgements

Special thanks to David Lazer for bringing our attention to Adam Bonica’s work, to David Parkes for suggesting a reformulation of our model that enabled integrating out voter positions, and to Matt Blackwell for encouraging us to think more about validity and identifiability. This work was supported in part by the NSF GRFP under grant #1122374. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect those of the sponsors.

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Alison Nahm .

Editor information

Editors and Affiliations

Appendices

A Mathematical Definitions of DiMaggio’s Polarization Metrics

Given the estimates of our model, we use the following analytical form of the standard deviation of a mixture model to measure political polarization in terms of dispersion:

$$\begin{aligned} M_{\sigma } = \sqrt{\left( \sum _{i=1}^K \frac{n_i}{\sum _{j=1}^K n_j} ({\mu _i}^2 + {\sigma _i}^2) \right) - M_\mu } \end{aligned}$$
(4)

where \(n_i\) is the total number of voters assigned to component i and \(M_{\mu }\) is the weighted mean of the mixture distribution of voter preferences.

To measure political polarization in terms of bimodality, we use kurtosis. Kurtosis is the fourth central moment of the mixture distribution divided by the square of the variance of the mixture distribution. We use the following analytical form:

$$\begin{aligned} M_k = \frac{E[(X - M_\mu )]^4}{M_{\sigma }^4} \end{aligned}$$
(5)

where X is a random variable drawn from the mixture distribution and hence the numerator is the fourth central moment of the mixture distribution. The analytical form to compute the z-th central moment of the mixture distribution is below.

$$\begin{aligned} E[(X - M_\mu )^z] = \sum _{i=1}^K \sum _{j=1}^z {z \atopwithdelims ()j} (\mu _i - M_\mu )^{z - j} w_i \, E[(Y_i - \mu _i)^z] \end{aligned}$$
(6)

where \(Y_i\) is a random variable drawn from component i of the mixture distribution, \(w_i\) is the weight of each component, and \(E[(Y_i - \mu _i)^z]\) is the z-th central moment of the ith component distribution. In our analysis, we weight each component in the mixture distribution by the proportion of the population assigned that component.

B Additional Results

In Sect. 5, we presented the results of our method assuming the underlying component distribution is Normal and the number of clusters (K) is 4. This section tests the robustness of these assumptions and presents our results when varying the underlying component distribution and the number of clusters.

1.1 B.1 Varying the Underlying Component Distribution

We test the inference procedure of our model not only assuming Normal component distributions, but also Uniform and Laplace component distributions. When assuming the distributions of voters follow a Laplace distribution, we use the same Normal prior defined for the mean of the Normal component for the location parameter and the same Inverse Gamma prior defined for the standard deviation of the Normal component for the scale parameter. When we use Uniform component distributions, we use the same Normal prior defined for the mean of the Normal component for both the minimum and the distance between the minimum and maximum parameters. The priors defined for the Normal component parameters can be found in Sect. 4. For each alternative underlying component distribution, the inferred distributions can be seen in Fig. 4, derived polarization metrics can be seen in Table 2, and prediction comparisons can be seen in Fig. 6.

Figure 5 visualizes the comparisons between the results derived of alternative component distributions and alternative data sources described in Sect. 5.1. We find significant positive correlations between our district-level point estimates and all of the alternative data sources. Assuming Laplace component distributions, our results have a correlation of 0.3216 with the responses selecting ideology given a discrete scale (left column in Fig. 5), 0.2514 with the responses selecting ideology along a continuous scale (middle column), and 0.5323 with the MRP estimates (right column). All of these correlations were significant with p-values less than 0.01. Assuming Uniform component distributions, our results have a correlation of 0.3652 with the responses selecting ideology given a discrete scale, 0.2404 with the responses selecting ideology along a continuous scale, and 0.6331 with the MRP estimates. Again, all of these correlations were significant with p-values less than 0.01.

1.2 B.2 Varying the Number of Clusters

We also varied the number of clusters (K) used in the model. The main results section in the paper presented results assuming \(K=4\), but below we include the inferred distributions in Fig. 7, derived polarization metrics in Table 3, and prediction comparisons in Fig. 9 for \(K=2\) and \(K=8\), assuming Normal underlying precinct distributions. Due to time constraints, we were only able to generate these results given the Texas and New York congressional elections.

Figure 8 visualizes the comparisons between the results derived of alternative component distributions and alternative data sources described in Sect. 5.1. We find significant positive correlations between our district-level point estimates and all of the alternative data sources. When our model assumes 2 clusters rather than 4 clusters, the results of our model have a correlation of 0.2845 with the responses selecting ideology given a discrete scale (left column in Fig. 5), 0.2646 with the responses selecting ideology along a continuous scale (middle column), and 0.7160 with the MRP estimates (right column). All of these correlations were significant with p-values less than 0.01. When our model assumes 8 clusters, the results of our model have a correlation of 0.2925 with the responses selecting ideology given a discrete scale, 0.1867 with the responses selecting ideology along a continuous scale, and 0.6861 with the MRP estimates (right column). Again, all of these correlations were significant with p-values less than 0.01.

Rights and permissions

Reprints and permissions

Copyright information

© 2016 Springer International Publishing AG

About this paper

Cite this paper

Nahm, A., Pentland, A., Krafft, P. (2016). Inferring Population Preferences via Mixtures of Spatial Voting Models. In: Spiro, E., Ahn, YY. (eds) Social Informatics. SocInfo 2016. Lecture Notes in Computer Science(), vol 10046. Springer, Cham. https://doi.org/10.1007/978-3-319-47880-7_18

Download citation

  • DOI: https://doi.org/10.1007/978-3-319-47880-7_18

  • Published:

  • Publisher Name: Springer, Cham

  • Print ISBN: 978-3-319-47879-1

  • Online ISBN: 978-3-319-47880-7

  • eBook Packages: Computer ScienceComputer Science (R0)

Publish with us

Policies and ethics