Summit will present a predictive analytics case study at the Federal Committee on Statistical Methodology (FCSM) 2015 Research Conference in Washington, DC.
FCSM conference attendees can join Summit's presentation on December 2, 2015, at 1:45pm in Room 146 C. (See program details here; Summit's presentation abstract is on page 63.)
From Summit's conference abstract:
Predictive Analytics with Administrative Data from the Mine Safety and Health Administration
Yuwen Dai (Summit Consulting, LLC), Natalie Patten (Summit Consulting, LLC), Albert Lee, Ph.D. (Summit Consulting, LLC), and George Cave, Ph.D. (Summit Consulting, LLC)
The U.S. Department of Labor (DOL), Mine Safety and Health Administration (MSHA) would like to predict at an early stage which mine operating firms with violations are at risk of failing to pay their fines on time. Summit Consulting, LLC (Summit), in collaboration with the DOL Chief Evaluation Office (CEO), used MSHA internal administrative data to develop predictive models that identify mine operators who are at high risk of failing to make timely payments to MSHA. Summit also created a dynamic, user-friendly Microsoft Excel tool (the Early Detection tool) based on the delinquency risk scores calculated by the predictive models.
The early identification of delinquent operators has three components: 1) identification of the probability to be delinquent, and 2) the severity of delinquency in terms of duration of delinquency, and 3) the compounding effect of delinquency probability and severity. Summit used logistic regression to predict whether a firm would be over 90 days delinquent, survival analysis to determine how long a firm would be in delinquency, and developed a composite risk score to rank the delinquency risk on violations based on both the predicted probability of delinquency and the duration of the delinquency. The early detection model has demonstrated good predictive power to discriminate operators with high delinquency risk and has uncovered key relationships in the delinquency behavior in a statistical and systematic way. Summit implemented the early detection model in a Microsoft Excel application so that MSHA can periodically use in-house resources to conduct the analysis and use the results for early enforcement activities. The highly automated tool requires minimum inputs and operations from users and has a user-friendly, dynamic interface.
This project increases MSHA’s internal capacity for conducting and implementing program-enhancing data analytics and serves as a proof of concept for using predictive analytics to improve agency performance.