This study aimed to answer the question of how media users will reallocate their sleep time when their main content channel changes from real-time broadcasting to selective viewing through the over-the-top (OTT) media streaming service. To draw a causal inference between OTT consumption and sleep patterns, the difference-in-difference (DID) estimation method was applied. With the DID approach, a clear distinction between treatment and control groups is essential because the main treatment effects can be screened by the compounding effects. While the conventional way of dividing two groups relies on the selection of limited variables, this study adopted random forest nearest-neighbor propensity score matching based on a machine learning algorithm to divide the two groups. This allows for meticulous matching of the two groups except for treatment. Results show that watching OTT late at night has a significant effect on reducing the total sleep duration on average by about 18–20 min (maximum about 30 min at 95% confidence level) and delaying bedtime by about 18 min (maximum about 26 min at 95% confidence level). This study showed that the selective viewing of content through OTT has the advantage of widening the range of content choices for media users and helping in arranging their time more autonomously, but watching content through OTT late at night leads to media users’ departure from the existing sleep routine.

本研究旨在回答以下问题：当媒体用户的主要内容频道从实时广播转变为通过过顶 (OTT) 媒体流服务进行选择性观看时，他们将如何重新分配他们的睡眠时间。为了得出 OTT 消耗和睡眠模式之间的因果关系，应用了差对差 (DID) 估计方法。使用 DID 方法，必须明确区分治疗组和对照组，因为可以通过复合效应筛选主要治疗效果。传统的两组划分方法依赖于有限变量的选择，而本研究采用基于机器学习算法的随机森林最近邻倾向得分匹配来划分两组。这允许在治疗之外对两组进行细致的匹配。结果表明，深夜观看 OTT 对平均减少总睡眠时间约 18-20 分钟（在 95% 置信水平下最多约 30 分钟）和延迟就寝时间约 18 分钟（在 95% 置信水平下最多约 26 分钟）有显着影响95% 的置信水平）。本研究表明，通过OTT选择性观看内容具有拓宽媒体用户内容选择范围的优势，有助于更自主地安排时间，但深夜通过OTT观看内容会导致媒体用户偏离现有的睡眠常规。