基于FII的真人秀投票机制公平性分析

Fairness Analysis of Reality Show Voting Mechanism Based on FII

摘要: 真人秀节目的投票机制是联结专业评判与观众偏好的枢纽，但观众投票数据的隐蔽性使公平性系统评估面临挑战。本文以《与星共舞》(DWTS) (2005~2024) 34个赛季421名参赛者为对象，提出可行区间反演(Feasible Interval Inversion, FII)方法，通过线性规划从淘汰结果反推观众投票份额的严格可行区间，无需预设先验分布。结果显示，FII方法对269个淘汰周的预测准确率达78.4%，优于随机猜测基准(42.1%)。排名制与百分比制的全赛季蒙特卡洛模拟比较表明，两者在74.3%的淘汰周中结论一致，但百分比制赋予观众的边际影响力显著更高。人口学分析揭示年龄与评委评分呈负相关(r = −0.398)，但与观众投票份额几乎无关(r = −0.059)，呈现出与怀旧溢价假说一致的模式(该关联的因果性有待纳入赛前知名度等控制变量后进一步检验)；运动员群体的中位生存周数比娱乐业选手少约1.5周。针对Bobby Bones和Bristol Palin等争议选手的个案分析确认了观众投票在极端情况下可完全覆盖评委判断。最后，本文提出了包含透明权重、异常值阻尼和评委挽救机制的改进投票系统，模拟表明其在47%的争议周中可将结果向专业评判方向校正。

Abstract: The voting mechanism of reality competition shows balances professional judgment and audience preference, yet the opacity of audience voting data challenges systematic fairness evaluation. Using records from 421 contestants across all 34 seasons of Dancing with the Stars (2005~2024), this study develops a Feasible Interval Inversion (FII) method that infers audience vote share feasible intervals from elimination outcomes via linear programming, requiring no prior distributional assumptions. Empirical results show 78.4% accuracy in predicting eliminations across 269 weeks, exceeding the random baseline of 42.1%. Monte Carlo comparison of rank-based and percentage-based vote aggregation reveals consistent predictions in 74.3% of weeks, with the percentage method conferring greater audience influence. Demographic analysis reveals a pattern consistent with a nostalgia premium hypothesis—age correlates negatively with judge scores (r = −0.398) but negligibly with audience votes (r = −0.059)—and athletes survive approximately 1.5 weeks less than entertainment professionals. Case studies of controversial contestants (Bobby Bones, Bristol Palin) confirm that audience voting can override professional judgment in extreme cases. An improved voting system with transparent weights, outlier dampening, and a judge-save provision is proposed, with simulation showing correction toward professional assessment in 47% of disputed weeks. (Causal attribution requires further validation controlling for pre-existing fame and media exposure).