Between Fall 2011 and Fall 2012 students at Utah State University played several rounds of Humans versus Zombies (HvZ), a role-playing variant of tag popular on college campuses. The goal of the game is for the zombies to tag humans, converting them into more zombies. Based on portrayals of ‘zombieism’ in popular culture, one might treat HvZ as a disease system. However, a traditional SIR model with mass-action dynamics does a poor job of modeling HvZ, leading to the natural question: What mechanisms drive the dynamics of the HvZ system? We use model competition, with Bayesian Information Criterion as arbiter, to answer this question. First, we develop a suite of models with a variety of transmission mechanisms and fit to data from fall 2011. We use model competition to determine which model(s) have the most support from the data, thereby offering insight into driving mechanisms for HvZ. Bootstrapping is used to both assess the significance of individual mechanisms and to determine confidence in the performance of our models. Finally, we test predictions of the best models with data from fall 2012. Results indicate that through both years of the game humans tend to cluster defensively, zombies tend to hunt in groups, some zombies are more proficient hunters, and some humans leave the game.

中文翻译：

---

28 模型之后：模型竞争和僵尸启示录

从 2011 年秋季到 2012 年秋季，犹他州立大学的学生玩了几轮人类大战僵尸 (HvZ)，这是一种在大学校园里流行的角色扮演游戏。游戏的目标是让僵尸标记人类，将他们变成更多的僵尸。根据流行文化中对“僵尸主义”的描述，人们可能会将 HvZ 视为一种疾病系统。然而，具有质量作用动力学的传统 SIR 模型在对 HvZ 建模方面做得很差，这导致了一个自然问题：什么机制驱动了 HvZ 系统的动力学？我们使用以贝叶斯信息准则为仲裁者的模型竞争来回答这个问题。首先，我们开发了一套具有各种传输机制并适合 2011 年秋季数据的模型。我们使用模型竞争来确定哪些模型获得数据的最大支持，从而深入了解 HvZ 的驱动机制。Bootstrapping 用于评估单个机制的重要性并确定对我们模型性能的信心。最后，我们使用 2012 年秋季的数据测试了对最佳模型的预测。结果表明，在这两年的游戏中，人类倾向于防御性地聚集，僵尸倾向于成群结队地狩猎，一些僵尸是更熟练的猎人，而一些人类则离开了游戏.