Multipolar dynamics of social segregation: Data validation on Swedish vaccination statistics

TL;DR

The paper addresses how to validate a multipolar opinion-dynamics model on real, correlative socio-behavioral data. It introduces a workflow that encodes predictor information as spatial bias distributions on a Watts-Strogatz graph and infers outcomes from the simulated final opinions, enabling comparison with aggregate and regional data. Applying this to Swedish SCB2022 data on vaccination uptake and political participation, the study demonstrates that spatially correlated biases are essential to reproduce both global trends and regional polarization, and that mixing biases can increase the penetration of the majority stance. The work highlights the importance of network topology in driving observed segregation, offers a framework for intervention exploration, and points to extensions with richer topologies and additional variables for broader applicability.

Abstract

We perform a validation analysis on the multipolar model of opinion dynamics. A general methodology for using the model on datasets of two correlated variables is proposed and tested using data on the relationship between COVID-19 vaccination rates and political participation in Sweden. The model is shown to successfully capture the opinion segregation demonstrated by the data and spatial correlation of biases is demonstrated as necessary for the result. A mixing of the biases on the other hand leads to a more homogeneous opinion distribution, and greater penetration of the majority opinion, which here corresponds to a decision to vote or vaccinate.

Figures (7)

• Figure 1: The dataset SCB2022 used for our analysis, dots represent regional statistical units
• Figure 2: Model output in the base case. The political participation rate is determined by the average of the first opinion index in each graph neighborhood corresponding to a regional statistical unit. Comparison to the measured data in Fig.\ref{['fig:scb']} shows a close resemblance of the overall shape.
• Figure 3: Comparing the histograms, we can see that the model skews slightly to the right. The close adherence to the measured outcome supports the validity of the model on an aggregate level. However, comparison with the more granular data presented in Fig.\ref{['fig:skane']} reveals that this does not hold locally. As discussed below, this local inconsistency is likely due to the approximate nature of the graph on the local level.
• Figure 4: When the same total amount of pro and anti-vaccination agents are distributed uniformly amongst the graph, the results stop resembling the target dataset. This highlights the importance of topology when using the multipolar model -- or in other words, the fact that spatially correlated biases $\mathbf{r}^i$ strongly affect the final opinion distribution.
• Figure 5: A comparison of the histograms of the prediction achieved by linear regression and the target dataset reveals that, while it does achieve a low RMSE, the distribution has many isolated peaks and seems to be overall concentrated more towards the mean of the dataset.