QAP
1. Types of network hypotheses
根据两个维度(group / node / dyad & X / Y),可以划分出六种类型的假说:
对于group level hypotheses:No particular statistical issues due to network context——使用常规的回归技术往往就可以处理这种情形。
对于node level hypotheses:类似于group level,但有时会遇到non-independent outcomes的情况——使用auto-regressive model:
$$
Y = b_0 + \rho WY+b_1X_1 + b_2X_2 + \cdots + \epsilon
$$
- WY(W为n×n矩阵,Y为n×1向量):其他nodes的因变量对于某一个node因变量的影响。
- W_ij表示从node i到node j的一种联系程度,可以用geodesic distance来操作化。
- 也就是说,W不一定要是adjacency matrix
- rho表示这种网络效应的整体强度。
(dyad-level hypotheses的情形最为复杂,我们最后讨论)
2. Permutation test
事实上我们忽略了两个依然可以导致传统统计推断工具失灵的问题:
- What if you do not have a sample?(有总体,没有样本)
- What if the number of cases is really small?(样本量小)
使用permutation test解决。
2.1 Example: wine expert
Suppose we have four wines: Argentine, Australian, French, and California.
- Alleged wine expert is supposed to identify which is which >>> After tasting, he says bottle 1 is Argentine, b2 is Australian, b3 is Californian and b4 is French >>> He gets 50% correct.