Fig 1.
(a) Exposure between agents u and v is attractive since is less than prescribed tolerance T, while influence between u and w is repulsive since duw is greater than the tolerance threshold T.
(b) The probability that two agents interact given their opinion difference, under different levels of exposure.
Fig 2.
An example assignment of group identification using clustering on a typical initial distribution of the population.
Two groups are identified and edge cases are assigned no group.
Fig 3.
Illustration of qualitative types of polarization achievable by the model, displaying positions of agents in each of the two initial groups, and their evolution for the first fifteen time steps of a simulation.
Vertical position is equivalent to the opinion position; i.e., position between 0 and 1 in opinion space. Groups are rectangular areas centered at the opinion mean of the group, with length representing the size of the group (not the opinion area they cover) hence a longer rectangle when the whole population is in one group (as in Fig 3a). Colour reflects the opinion value too. Four qualitative polarization types are illustrated: (a) consensus, (b) stable polarization, (c) unstable polarization due to group drift, and (d) unstable polarization due to group fragmentation.
Fig 4.
Group-Dependent Tolerance (GDT) simulations.
The most important factor in determining polarization outcomes of the population is as it decides whether two groups will be attracted to each other or separate. This is dependent on E being high enough that out-group opinions are observed often enough to impact the in-group.
Fig 5.
Group-Dependent Responsiveness (GDR) simulations.
The size of response in attraction or repulsion governs a more gradual change of behavior when compared to the sharp boundaries for T in Fig 4. Out-group treatment is again most important when exposure to different opinions occurs regularly.
Fig 6.
Group-Dependent Exposure (GDE) simulations.
Over-exposure to in-group opinions and under-exposure to out-group is a scenario that mirrors concerns around filter bubbles. In our model, polarization is more likely with higher implying that polarization occurs from exposure to differing opinions that is not tempered by either high tolerance of opinions or larger exposure to in-group.