Reviews: Identification and Estimation of Causal Effects from Dependent Data

The work presented in the paper is clearly of value. The existing theory for identification and estimation of causal parameters in DAGs for IID data has been central to our understanding of causal inference, and developing analogous results for data under interference would be useful both to apply directly to data in which we know interference occurs and to better understand the potential impacts of violations of the IID assumption. While the paper should be accepted, the current version could be substantially improved in both organization and in its discussion of several key issues, including generality, assumptions, temporal effects, and prior work. Organization and Presentation Some aspects of the organization make the paper challenging for readers. Some sections do not provide a "roadmap" to the basic logic before plunging into the details, others do not present a high-level intuition for why a given theoretical result is being presented, The entire paper would be substantially improved if the authors provided readers with a high-level roadmap to the overall reasoning of the paper, making clear the basic logic that allows an identification theory to be developed under interference (before plunging into the details of sections 2 and 3). As I understand it, the outline of that logic is: 1) Represent models as latent variable chain graphs, 2) Divide the model into blocks, 3) Assume no interference between blocks, 4) Express identification theory by using truncated nested factorization of latent projection ADMGs.