Valid Statistical Inference for Network Dependent Data
Abstract
Project Summary/Abstract Many aspects of networks are active areas of research, but there is a glaring hole in our ability to learn about network systems, namely a dearth of valid methods for statistical inference about observations sampled from network nodes. Statistical inference is usually based on the assumption that the observations in a sample are independent with respect to the trait under study, but this assumption does not hold when the observations are linked to one another by social network ties and the trait of interest is affected by or related to those ties. New statistical methodology is needed in these settings. Broadly, we propose a two-pronged approach to the development of statistical methods for network data: rigorous, theoretical solutions, on the one hand, and accessible, easily implemented tools on the other. Aim 1 is to explore conditions under which central and other limit theorems hold for observations on individuals interconnected by network ties. Aim 2 is to develop hypothesis tests to detect network dependence, and Aim 3 is to develop ad hoc, effective-sample-size based corrections for network dependence.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 12, 2016
- Source ID
- N000141512343
Entities
People
- Elizabeth Leigh Ogburn
Organizations
- Johns Hopkins University
- Office of Naval Research
- United States Navy