Classic data structures that economists and statisticians have traditionally worked with are cross-sectional data, time series, and panel data. The main statistical tools for these types of data were developed in the 20th century, often in response to new kinds of information becoming available. For example, stock market data, national accounts, household surveys, and longitudinal studies of families each gave rise to new methods.
Over the past two decades the amount and variety of data in the social sciences have grown very quickly. Much of this growth is due to administrative records, digital traces, and new measurement technologies, together with deliberate efforts to collect larger and richer datasets. Many of these modern datasets cannot be described simply as cross sections or panels. Instead, they often have a network structure, linking individuals, firms, or countries to one another. In such settings the precision of statistical inference depends on the shape of the network itself.
The objective of the project was to build a robust statistical framework for these modern forms of data. The project aimed to establish a mathematical way of representing the structure of the data, to connect that structure to the reliability of statistical inference, and to design new methods that correct bias and produce more reliable standard errors.
Conclusion: The project has delivered on these aims. It has produced a set of new identification strategies, bias correction methods, and robust inference tools that make the analysis of complex panel and network datasets more trustworthy and useful for applied researchers.
