Source
Friedkin, Jia & Bullo, LIT-1482, read in full in NOTE-1126 from the version of record: pp. 452–471, Tables 2–6 and Figures 4–13, with supplement Sections S1–S4.
What was actually shown
120 undergraduates in groups of four discussed 15 choice dilemmas each. After every discussion each member split 100 chips between how far the discussion had and had not influenced their final opinion, and the first pile among the others. That gives each member’s closure (the second pile) and, through the Friedkin–Johnsen model, each member’s influence centrality on the outcome.
- The individual step. Prior centrality predicted closure on the next issue (Table 2, n = 1,680). The effect of time-averaged centrality grew along the sequence. The chance of complete closure rose with it (Fig. 6).
- Structure. Discussion time fell by about 0.9 minutes per issue (Table 4). Network density fell from 84% to 67% by issue 11 (Fig. 11). The dominant eigenvalue of the influence matrix and its trace declined (Figs 10, 13).
- Not shown. “In our data, the probability of reaching consensus does not alter along the issue sequence” (p. 467). About a third of discussions each ended in disagreement, consensus on an initial opinion, and compromise.
What could have come out otherwise: closure on the next issue could have been unrelated to prior influence, as a fixed-personality view expects, and networks could have densified with interaction, as the usual sociological expectation has it. Neither happened.
What this does not say
- That reflected appraisal is the cause. Closure and centrality come from one self-report, closure persists across issues, and Table 2 does not control for prior closure. Stable individual differences could produce part of the association. Nothing manipulated perceived influence.
- That groups become autocracies. That is the model’s attractor (all 14,000 simulated systems, S1) and the authors’ extrapolation. The data give “signals”.
- That the dynamics are DeGroot–Friedkin. The single-issue model here is Friedkin–Johnsen, with anchoring. THEORY-916 holds the DeGroot version’s theorems, whose end state differs.
- That concentration is bad or good. The authors decline to say whether the efficiency gained goes with more or less collective intelligence (S11).
Connections
- CLAIM-448. Supports its baseline, self-reinforcing authority, with data, and its point that faster settlement is not better integration. The thinning is of uptake: members stop weighting some others, rather than a regime keeping contests off the record (CLAIM-403’s ladder). The supplement’s untested “reserved authority” argument complicates the claim’s prediction for a protected challenge channel.
- THEORY-916. The same mechanism under DeGroot dynamics, proved rather than measured.