Statistical Methods for Evaluating the Relationships Among Social, Spatial and Genetic Networks
Abstract
Empirical findings suggest close relationships among social, spatial and genetic networks. Information learned from one network would provide beneficial insight on the others. These relationships offer the flexibility to obtain understanding of networks that are hard to observe, such as ones for disease transmission.In Chapter 1, we start from theoretical models and empirical findings which suggest that the intensity of communication among groups of people declines with their degree of geographical separation. Based on the evidence that rather than decaying uniformly with distance, the intensity of communication might decline at different rates for shorter and longer distances, we introduce a statistical model based on Bayesian LASSO for estimating the rate of communication decline with geographic distance that allows for discontinuities in this rate. We apply our method to an anonymized mobile phone communication dataset and discover some geographic patterns.
Based on the findings in Chapter 1, it becomes clear that methods which can provide statistical justification for association between communities based on different networks are important to infer connections in one network based on another network. In Chapter 2, we justify the use of a permutation test focusing on testing the null that there is no association between pairs of nodes in the same community in one assignment and they are in the same community according to the other assignment. In simulation, we evaluate its performances and find that the permutation test preserves the type I error and has adequate power. The use of the test is then demonstrated on work and social relations data of a tailor shop in Zambia.
In Chapter 3, with the tools introduced in Chapter 2, we take one step closer to evaluate association between clusters based on geographic distances and on viral genetic distances. By applying the hypothesis testing framework to analyze this association among individuals infected by HIV viral strains, we find significant association between geographic pattern and viral genetic clusters, which is not greatly affected by different divergence threshold values used in the construction for viral genetic clusters. Our findings provide an alternative view and are consistent with previous research.
Terms of Use
This article is made available under the terms and conditions applicable to Other Posted Material, as set forth at http://nrs.harvard.edu/urn-3:HUL.InstRepos:dash.current.terms-of-use#LAACitable link to this page
http://nrs.harvard.edu/urn-3:HUL.InstRepos:40050129
Collections
- FAS Theses and Dissertations [6138]
Contact administrator regarding this item (to report mistakes or request changes)