M1-085M1: Mathematics & Statistics Fundamentals实验设计 (A/B)Hard
⚡Flashcard Drill➔Mastery:

Explain Why Individual Randomization Fails Under Network Spillover/Interference and How Cluster Randomization Solves It

📐 Mathematical Definition
SUTVA: Yi(Ti) 独立于 Tj (j≠i)\text{SUTVA}:\ Y_i(T_i)\ \text{独立于}\ T_j\ (j\ne i)
⚡ One-Sentence Core Answer
Individual randomization fails under network interference because treating a user spills over to affect their connected social peers, violating SUTVA and contaminating control group baseline measurements.

📌 Key Takeaways

  • •
    SUTVA Violation: In social networks (Meta, LinkedIn) or collaborative platforms (Slack, Figma), YiY_i depends not just on TiT_i, but on the treatment vector of friends TN(i)T_{\mathcal{N}(i)}.
  • •
    Direction of Bias: Direct network effects typically cause Treatment to elevate Control engagement (e.g. Chatting with untreated friends), severely diluting and underestimating the true treatment lift.
  • •
    Primary Mitigations: Graph Cluster Randomization (assigning tightly connected subgraphs to the same variant) and Ego-Network Randomization.

📐 Mathematical Derivations

Network potential outcomes model (Manski, 2013): Let Yi(t,s)Y_i(t, s) depend on individual treatment t∈{0,1}t \in \{0, 1\} and network exposure s=∑j∈N(i)Tj∣N(i)∣∈[0,1]s = \frac{\sum_{j \in \mathcal{N}(i)} T_j}{|\mathcal{N}(i)|} \in [0, 1]. Under individual Bernoulli randomization, control users are exposed to s>0s > 0. The observed difference in group means is: τ^naive=E[Yi(1,s)∣Ti=1]−E[Yi(0,s)∣Ti=0]\hat{\tau}_{\text{naive}} = E[Y_i(1, s) \mid T_i=1] - E[Y_i(0, s) \mid T_i=0]. However, the true total launch effect (Global Treatment Effect) is τglobal=E[Yi(1,1)]−E[Yi(0,0)]\tau_{\text{global}} = E[Y_i(1, 1)] - E[Y_i(0, 0)]. If positive spillovers exist (E[Y(0,s)]>E[Y(0,0)]E[Y(0, s)] > E[Y(0, 0)]), then τ^naive\hat{\tau}_{\text{naive}} systematically underestimates the true global product value.

🏭 Production Trade-offs

Cluster Randomization protocol: (1) Run graph partitioning algorithms (Louvain or spectral clustering) to segment the social graph into isolated clusters minimizing cut edges between clusters. (2) Randomize entire clusters to Treatment or Control. This ensures friends share the same treatment status, containing spillovers. Trade-off: Reduces effective degrees of freedom from millions of individual users down to the number of clusters KK, drastically inflating variance and requiring cluster-robust standard errors.
⚠️ Common Interview Pitfalls
  • ✕
    Using individual-level standard errors on cluster-randomized data (drastically inflates false positive rate; must cluster standard errors by cluster ID).
  • ✕
    Allowing large inter-cluster boundary edge cuts, which lets spillovers bleed across cluster boundaries.
🎯 Interviewer Follow-up Inquiries
  • ?
    How does the Exposure Model (Aronow & Samii, 2017) decompose network effects into direct and indirect spillover effects?
  • ?
    Why does graph modularity optimization (Louvain) maximize experimental power in network experiments?
⚡
Test muscle memory on this topic via 3-Layer Flashcards
Active Recall: Mentally formulate Prompt ➔ Math Derivation ➔ Benchmark Implementation in 3 minutes.
🎯 Drill This Card Now➔
📚

Associated Knowledge Base Guides & Mindmaps

Explore the comprehensive technical article, exam cards, and global architecture tree.

← PreviousM1-084: Explain Switchback Experiments and Their Applications in Marketplace and Two-Sided Delivery Platforms📋Back to BankNext →M1-086: Explain Trigger Analysis in Online Experiments and How It Enhances Statistical Power
中文速览与面试要点

解释网络效应/干扰下为什么个体随机化会失效,以及如何应对。

个体处理会影响他人(溢出),违反 SUTVA;用集群随机化、switchback 或专门的干扰感知设计。

核心要点

  • 干扰的机制:标准 A/B 要求 SUTVA (个体 i 的结果只依赖自己的处理 Tᵢ),但在以下场景被违反——① 社交网络 :处理组用户的行为会影响其好友(如分享、跟风),使对照组也被'间接处理';② 双边市场 :买家实验改变需求,影响卖家行为与另一侧用户体验;③ 共享资源 :处理组占用更多库存/运力,挤压对照组(如打车平台的补贴实验);④ 竞争/学习 :同一关键词的广告预算实验会互相竞价。 后果 :个体随机化测到的是'直接效应 + 部分溢出效应'的混合——若溢出是 正的 (对照组也受益),则处理效应被 低估 ;若是 负的 (对照组被挤压),则被 高估 。且 干扰强度越大、偏差越大 ;严重时甚至符号翻转。
  • 应对策略:① 集群随机化(Cluster Randomization) ——以'无干扰的单元'为随机化单位(如地理区域、社交社区、城市);优点是保留个体层面数据(功效高于 switchback);代价是 有效样本量降为集群数 (集群内相关导致方差膨胀,需用 ICC(组内相关系数) 校正:设计效应 = 1+(m−1)·ICC),且集群数需足够(通常 ≥40)。② Switchback(时间轮换) ——把干扰限制在时间窗内(见上题);适合无法按空间划分的场景。③ 饱和设计(Saturation Design) ——刻意让部分集群的处理比例不同(如 0%、50%、100%),通过对比不同饱和度的集群 估计溢出效应 ;这是 Facebook 研究社交影响的标准方法。④ 双向随机化(Bipartite/Two-sided) ——对市场两侧分别随机化并联合分析。⑤ 检测干扰 ——比较不同饱和度/集群大小的效应估计(若随饱和度变化则存在干扰);或用'影子对照'(未暴露于任何处理的全隔离组)。⑥ 实践建议 ——先评估干扰强度(小规模集群实验),若干扰弱可用个体随机化 + 稳健标准误,若干扰强必须用集群/switchback;且应报告'干扰下的效应解读'(是直接效应还是总体效应)。
  • 干扰导致效应被稀释或放大(方向不定)
  • 干扰越强,个体随机化的偏差越大

常见陷阱

  • 在有网络效应的场景用个体随机化
  • 集群随机化忽略 ICC 导致方差低估