Interesting progression of HAC / cluster-robust inference, from serial correlation, to spatial correlation (e.g., Müller and Watson, 2021, here), and now network-induced correlation.
Very cool new network result from Michael Leung at USC: Asymptotic independence (in the number of linking steps), the key to robust/clustered inference in network environments, holds iff network clusters have conductance (the ratio of edge boundary size to volume) approaching 0. (Yes, necessary and sufficient!)
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.