Interest in the collection and analysis of dynamic network data has increased dramatically over the last decade. This growth coincides with technological developments, such as computational resources and the Internet, that allow for improved measurement, collection, and modeling of inter-temporal network data. Examples of social networks include structures of friendships, job leads, or emergency information ties among individuals; joint ventures among firms; and alliances among nations. Modern data collection through sensors (e.g., cellphones), surveys, and database systems has allowed for larger and more detailed dynamic network data collection than was possible in past decades; however, even with improved measurement tools there exists a persistent problem of missing data, either by design (e.g., sampling) or out of design (e.g., machine failure). Thus, the collection of data on large dynamic networks often results in missing data, which requires new methodology for estimation and simulation. This doctoral dissertation research project will employ computational methods, exponential family theory, and a latent missing data framework to develop models that will be evaluated with real-world empirical cases. The project consists of several linked activities. The research will extend current missing data techniques employed in the statistical analysis of social network data to the context of dynamic networks with and without vertex dynamics. Several competing likelihood-based missing data methods under the framework of multiple imputation will be developed. In addition, the research will evaluate these missing data models through a series of simulation experiments to compare the efficiency, scalability, bias, accuracy, and predictive accuracy of these missing data techniques.
This project will improve and extend the current state of the art in missing and sampled data methods for dynamic network models. These methods will allow improved inference and prediction for dynamic social network processes (e.g., online social networks, disaster response networks, etc.), problems of immediate importance to sociologists, statisticians, computer scientists, demographers, epidemiologists, and public policy researchers. The test cases used within this research are drawn from real-world cases of interest to the greater public, so these methods should enhance the work of practitioners in hazards research, public health, and public policy. As a Doctoral Dissertation Research Improvement award, support is provided to enable a promising student to establish a strong, independent research career.
The collection and analysis of dynamic network data has increased dramatically over the last decade. This growth coincides with technological developments, such as computational resources and the Internet, that allow for improved measurement, collection, and modeling of inter-temporal network data. Examples of social networks include structures of friendships, job leads, or emergency information ties among individuals; joint ventures among firms; and alliances among nations. Modern data collection through sensors (e.g., cellphones), surveys, and database systems has allowed for larger and more detailed dynamic network data collection than was possible in past decades; however, even with improved measurement tools there exists a persistent problem of missing data, either by design (e.g., sampling) or out of design (e.g., machine failure). Thus, the collection of data on large dynamic networks often results in missing data, which requires new methodology for estimation and simulation. This doctoral dissertation research project employed computational methods, exponential family theory, and a latent missing data framework to develop models that were evaluated with real-world empirical cases. The research extended current missing data techniques employed in the statistical analysis of social network data to the context of dynamic networks with and without vertex dynamics. Several competing likelihood-based missing data methods under the framework of multiple imputation were developed. In addition, the research evaluated these missing data models through a series of simulation experiments to compare the efficiency, scalability, bias, accuracy, and predictive accuracy of these missing data techniques. This results of this project improved and extended the current state of the art in missing and sampled data methods for dynamic network models. These methods allow improved inference and prediction for dynamic social network processes (e.g., online social networks, disaster response networks, etc.), problems of immediate importance to sociologists, statisticians, computer scientists, demographers, epidemiologists, and public policy researchers. The test cases used within this research are drawn from real-world cases of interest to the greater public, so these methods should enhance the work of practitioners in hazards research, public health, and public policy. As a Doctoral Dissertation Research Improvement award, support was provided to enable a promising student to establish a strong, independent research career; this grant directly contributed to Co-PI’s completion of his doctoral dissertation and employment as a tenure track assistant professor in Statistics and Sociology.
