How social interactions mitigate extremist views

Soci­ety is becom­ing increas­ingly divided, and we are con­tinu­ing to see the emer­gence of extrem­ist views around the world, be it with regards to top­ics like polit­ics, reli­gion, or cli­mate change. While there has been a great deal of research into how this phe­nomen­on, which is known as ‘polar­isa­tion’, has evolved less atten­tion has been paid to under­stand­ing how social inter­ac­tions can lead to the oppos­ite effect – “depol­ar­isa­tion” –, which occurs when indi­vidu­als begin to modi­fy their opin­ions so that they are less extreme.

To address this ques­tion, Jaume Ojer, Michele Star­n­ini and Romu­aldo Pas­tor-Sat­or­ras, from the Departa­ment de Física, Uni­versit­at Politèc­nica de Catalun­ya and the CENTAI Insti­tute in Tur­in, have pro­posed a new “social com­pass” mod­el to study how opin­ion var­ies between groups with extrem­ist pos­i­tions and how these opin­ions might be depol­ar­ised¹. Their the­or­et­ic­al frame­work has been val­id­ated by extens­ive numer­ic­al sim­u­la­tions and tested using data from opin­ion polls col­lec­ted by the Amer­ic­an Nation­al Elec­tion Studies.

“Polar­isa­tion may con­trib­ute to widen­ing the polit­ic­al divide in our soci­ety, ham­per­ing the col­lect­ive res­ol­u­tion of import­ant soci­et­al chal­lenges,” say the research­ers. “It could even encour­age the spread of mis­in­form­a­tion and con­spir­acy the­or­ies. Our depol­ar­isa­tion frame­work could provide solu­tions to these soci­et­al ills.”

Mod­els describ­ing polar­isa­tion are based on mech­an­isms as diverse as homo­phily, bounded con­fid­ence or opin­ion rejec­tion. Until now, the depol­ar­isa­tion pro­cess in a pop­u­la­tion has gen­er­ally been mod­elled for the simple case of an indi­vidu­al’s opin­ion on a single sub­ject. In real­ity, how­ever, an indi­vidu­al gen­er­ally has opin­ions on sev­er­al sub­jects at any giv­en time. A mul­ti­di­men­sion­al mod­el­ling frame­work is there­fore needed to bet­ter describe how opin­ions evolve.

When mul­tiple sub­jects are con­sidered, a num­ber of fea­tures emerge. The first is align­ment, that is, the pres­ence of a cor­rel­a­tion between opin­ions with respect to dif­fer­ent sub­jects. For example, people with strong reli­gious con­vic­tions are more likely to oppose abor­tion legis­la­tion. The prob­lem with cur­rent mul­ti­di­men­sion­al mod­els is that they neg­lect this inter­de­pend­ence between dif­fer­ent sub­jects, which means that they fail to clearly describe opin­ion polarisation.

The key idea of the social com­pass mod­el is to rep­res­ent opin­ions in rela­tion to two top­ics loc­ated on oppos­ite sides of a polar plane. The angle of the plane rep­res­ents an indi­vidu­al’s ori­ent­a­tion as regards to the two top­ics, and its radi­us expresses the strength of the atti­tude (or ‘con­vic­tion’).

“This polar rep­res­ent­a­tion nat­ur­ally allows us to for­mu­late the key hypo­thes­is of our mod­el, namely that intransigents with extreme opin­ions (or strong con­vic­tion) may be less likely to change their opin­ion than indi­vidu­als with weak con­vic­tion,” explains Michele Star­n­ini. This hypo­thes­is is intu­it­ive and con­sist­ent with obser­va­tions made in exper­i­ment­al psy­cho­logy. “Such a polar rep­res­ent­a­tion is very com­mon in phys­ics, but not so much in the social sciences.”

Inspired by the Friedkin-Johnsen mod­el², the research­ers stud­ied how social influ­ence can affect the ini­tial opin­ions of indi­vidu­als. They found that their mod­el describes a phase trans­ition from an ini­tial polar­ised state to a depol­ar­ised state as a func­tion of increas­ing social influ­ence. The nature of this trans­ition depends on the dis­par­ity of ini­tial opin­ions: opin­ions that diverge strongly at the out­set trig­ger a so-called first-order (or explos­ive) depol­ar­isa­tion towards con­sensus, while opin­ions that are more cor­rel­ated to begin with lead to a second-order (or con­tinu­ous) transition.

To test their mod­el, the research­ers used data on cor­rel­ated top­ics – such as abor­tion and reli­gion – and uncor­rel­ated top­ics – for example, immig­ra­tion and mil­it­ary dip­lomacy in the United States – from the Amer­ic­an Nation­al Elec­tion Stud­ies. They found that com­munit­ies asked to give their opin­ion on these sub­jects under­went a phase trans­ition from polar­isa­tion to depol­ar­isa­tion in numer­ic­al sim­u­la­tions of the mod­el, with indi­vidu­als in the com­munity inter­act­ing and influ­en­cing each other.

They stud­ied the mod­el under “mean field” con­di­tions, mean­ing that each indi­vidu­al can inter­act with all the oth­er indi­vidu­als. “Since opin­ions are described by angles, it was nat­ur­al for us to mod­el con­sensus form­a­tion as the align­ment of agents’ ori­ent­a­tions,” explains Michele Star­n­ini. “This type of phase coup­ling is inspired by the Kur­amoto mod­el and is real­ist­ic for small groups. In future work, we will test our mod­el on large inter­act­ing groups, such as social networks.”

“Anoth­er inter­est­ing applic­a­tion that we are look­ing for­ward to imple­ment­ing involves sim­ul­tan­eously meas­ur­ing the opin­ions of indi­vidu­als with respect to mul­tiple top­ics and their social inter­ac­tions, to test the mod­el in this more real­ist­ic setting.”

Isabelle Dumé