针对2026年世界棒球经典赛(WBC)淘汰赛阶段的总得分预测,实际结果为59分1。此前Metaculus社区的预测均值接近70分,范围在60至80分之间;文章作者初始给出的预测值为60分,置信区间为42至80分1。复盘显示,这一结果验证了基于负二项分布的统计模型相较于简单基准模型更具准确性1。
在建模过程中,简单的历史数据平均值方法假设每场比赛平均得分约为8.7分(由61除以7得出),导致初步预测范围较宽,达35至87分1。然而,数据分析发现每队每局的得分并非独立事件,其方差大于均值,呈现出过度离散计数特征,因此不适合使用泊松分布进行拟合1。为解决这一问题,研究者选用参数为NB(μ=4.3, r=3.4)的负二项分布来模拟单队单局数据1。利用恒等式 sum(NB(μ, r)) ~ NB(nμ, nr),计算包含14个样本(7场比赛乘以2支球队)的和,最终得出的预测分布为NB(μ=60.2, r=47.6),对应的90%置信区间即为42至80分1。
The actual total score for the knockout stage of the 2026 World Baseball Classic (WBC) was 59 runs 1. While the Metaculus community prediction mean hovered near 70 with a range of 60-80, an individual analyst initially forecasted 60 runs within a broader range of 42-80 1. A simple baseline model, assuming an average of approximately 8.7 runs per game based on historical data (61/7), yielded a preliminary prediction range of 35-87 1. However, this basic approach failed to account for the statistical characteristics of baseball scoring distributions.
Upon analyzing the data, it became clear that scores were not independent events but exhibited overdispersion, where the variance exceeded the mean 1. To address this, the negative binomial distribution was selected to model single-team, single-inning data, fitted with parameters NB(μ=4.3, r=3.4) 1. By leveraging the additive property of the negative binomial distribution, specifically the identity sum(NB(μ, r)) ~ NB(nμ, nr), the analyst calculated the combined distribution for 14 samples, representing two teams across seven games 1. This refined modeling resulted in a final predicted distribution of NB(μ=60.2, r=47.6) 1. The resulting 90% confidence interval of 42-80 runs proved more accurate than the baseline estimate, successfully bracketing the actual outcome of 59 runs 1.
评论
还没有评论,欢迎留下第一条。