Collaborative Research: Accounting for geolocation error in point pattern analysis

Researchers in the social sciences increasingly utilize event data sets when studying crime, protests, and terrorism. These data sets provide information on each incident, including where it occurred, who was involved, what the consequences were, etc. Unfortunately, the recorded location of incidents in these data sets are often inaccurate, due to limitations in the available information from which they are drawn (ex. incomplete media reports). Left unaddressed, these geolocation errors impair one’s ability to effectively learn about the underlying process of interest from these data. For example, geolocation errors may cause researchers to infer spatial patterns from these data that would not be found with the correct locations. In this research, investigators will develop statistical methods to better account for geolocation errors in these kinds of data. The statistical methods developed will then be applied to data on political violence, demonstrating their importance for improved understanding of real-world problems. The multidisciplinary project will also provide training for the next generation of researchers at the intersection of statistics and the social sciences. This collaborative project includes support and mentorship for graduate students. Spatial point processes are a natural approach for modeling event data. However, geolocation errors produce two distinct, but related, problems for these methods: i) duplicate event locations, and ii) inaccurate spatial coordinate information. In this project, investigators will address both issues, developing a computationally efficient statistical inference method to account for geolocation error in spatial point pattern data within the Log-Gaussian Cox Process framework. Various geolocation error structures will be considered, including nonstationary errors, to better reflect complex real-world applications. The project will include research on both the finite-sample performance and asymptotic behavior of the estimators from the developed inference methods. These methods will be used to analyze real-world political violence data from various sources. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.