Synopsis
Rutgers University will extend formal tools of data science, adapting techniques from ecology and operations research. The resulting insights into missed detections will help CBP and other agencies assess and document performance; get early warning of change; assess trends in a timely fashion; and understand the effect of specific resource allocations on deterrence and detection.
The research will use synthetic data on interdictions, recidivism, and on effort, by station. The two novel methods proposed to exploit these data are: Extended/multi-type models of the Capture-Recapture concept (ECR), both passive and active, and the optimization techniques of Data Envelopment Analysis (DEA).
Simple capture-recapture models have already been applied in this context, and underlie some Border Patrol metrics. The currently employed “naïve” Capture-Recapture models assume that all apprehended persons try again, and are as likely to be caught as any other person. This project will develop more sophisticated ECR models of this complex process, define the data needed to apply it, and validate it with both simulated and actual data.
The DEA method arose in governmental and non-profit settings where multiple “Decision Making Units” (DMUs) deal with similar problems. DEA recognizes that each DMU, such as a border station or sector, differs from other DMUs. Using mathematical techniques including Linear Programming, DEA provides a principled way to assess resources and their relations to impacts. The management benefit of DEA is that each unit can be considered in its own context and with its own specific mix of resources and impacts. The engineering and scientific challenges lie in adapting models to the peculiarities of the border security problem, and in dealing with practical operational limits on the data that are, or can be, available to decision-makers.
Presentations and Reports
Presentation | 2017 PI Meeting
Poster | 2017 BTI Institute Showcase
Technical Report | Report on Deterrence: Summary and Sources
Technical Report | Report on Selected Data Sources
Technical Report | Report on Aggregating Imperfect Estimates of several key variables
Technical Report | Estimating Missed Detections by Combining Diverse 1 Sources: Algorithms Missed Detections Working Report-002
Technical Report | Estimating Missed Detections: A "Two Types" Model. Missed Detections Working Report ECR-003